Dynamic Control of District Heating Networks with Integrated Emission
Modelling: A Dynamic Knowledge Graph Approach
Markus Hofmeistera,b,c, Kok Foong Leeb, Yi-Kai Tsaib, Magnus Müllerb, Karthik Nagarajanb,
Sebastian Mosbacha,b,c, Jethro Akroyda,b,c, Markus Krafta,b,c,d,∗
aDepartment of Chemical Engineering and Biotechnology, University of Cambridge, Philippa Fawcett Drive, Cambridge, CB3 0AS, United
Kingdom
bCambridge Centre for Advanced Research and Education in Singapore, CARES Ltd., 1 Create Way, CREATE Tower #05-05, 138602, Singapore
cCMCL Innovations, Sheraton House, Cambridge, CB3 0AX, United Kingdom
dThe Alan Turing Institute, 96 Euston Road, London, NW1 2DB, United Kingdom
Abstract
This paper presents a knowledge graph-based approach for the dynamic control of a district heating network with
integrated emission dispersion modelling. We propose an interoperable and extensible implementation to forecast
the anticipated heat demand of a municipal heating network, minimise associated total generation cost based on a
previously devised methodology, and couple it with dispersion simulations for induced airborne pollutants to provide
automatic insights into air quality implications of various heat sourcing strategies. We create cross-domain interoper-
ability in the nexus of energy and air quality via newly developed ontologies and semantic software agents, which can
be chained together via The World Avatar dynamic knowledge graph to resemble the behaviour of complex systems.
Furthermore, we integrate the City Energy Analyst into this ecosystem to provide building-level insights into energy
demand and renewable generation potential to foster strategic analyses and scenario planning. Underlying calcula-
tions use building and weather data from the knowledge graph in place of inherent assumptions in the official software
release, facilitating a more data-driven approach. All use cases are implemented for a mid-size town in Germany as
a proof-of-concept, and a unified visualisation interface is provided, allowing for the examination of 3D buildings
alongside their corresponding energy demand and supply time series, as well as emission dispersion data. With this
work, we outline the potential of Semantic Web technologies to connect digital twins for holistic energy modelling in
smart cities, thereby addressing the increasing complexity of interconnected energy systems.
Keywords: knowledge graph, digital twin, interoperability, energy modelling, emission dispersion
1. Introduction
Climate change arguably poses humanity’s most
formidable challenge, impacting almost every aspect
of our lives [1, 2]. Recognising greenhouse gas emis-
sions as its key driver, the transition towards a low-
carbon future is widely acknowledged as a crucial im-
perative [3, 4]. The decarbonisation of the energy sector
requires significant changes, such as increased sector-
coupling and greater penetration of distributed renew-
able resources as well as the development of intelligent
infrastructure and modelling approaches [5, 6, 3]. So-
lutions for this inherently interdisciplinary transition re-
∗Corresponding author
Email address: mk306@cam.ac.uk (Markus Kraft)
quire holistic consideration of social, economic, envi-
ronmental, and engineering factors across various geo-
graphic and temporal scales [7].
Digital technologies like advanced metering infras-
tructure, big data, machine learning, 5G, and the inter-
net of things are increasingly recognised for facilitating
cost-effective decarbonisation [8, 9]. The application
of cyber-physical systems in energy research has grown
significantly in recent years, often in the form of digital
twins to explore optimal solutions for real-world prob-
lems through the study of fully digital replicas [10, 11].
Digital twins can provide detailed digital representa-
tions of assets, processes, or entire systems, describing
their current state and how they behave over time and
under different conditions and constraints. Digital twins
have effectively addressed numerous real-world prob-
Preprint submitted to Energy and AI May 8, 2024
lems [12]; however, the majority remains isolated from
each other and lack interoperability due to differences
in set-up, hardware or software, often stemming from
individual funding initiatives or business interests [13].
Interoperability is defined as the ability of tools, sys-
tems, and data to understand and use each other’s func-
tionalities [14], and is essential to foster reusability
of data and software assets and address cross-domain
questions comprehensively and collectively [15]. For
future energy systems, effective cooperation and coordi-
nation beyond the ‘traditional’ energy sector are essen-
tial to maximise synergies of increasingly intertwined
systems, encompassing power generation, the built en-
vironment, health, etc. A potential solution to this chal-
lenge can be generalised in the form of connected digital
twins - distributed collaborative entities that share data
and computational capabilities to efficiently and effec-
tively address complex questions [16].
It is anticipated that energy modelling will transition
from single-institution models to distributed, collabo-
rative approaches, allowing multiple domain experts to
contribute [17]. Yet, integrating data across domains
and resolving ambiguities while ensuring openness and
transparency remains a widespread problem [18, 19].
Data are highly heterogeneous in both format and se-
mantics, as different sources (i.e., sensors, texts, web,
etc.) use individual formats (e.g., tabular data, geospa-
tial data, natural language, etc.) [20]. Moreover, a
lack of semantic interoperability can arise when cer-
tain information is only known implicitly by domain
experts or when the same concept might possess differ-
ent meanings in different domains; however, an aligned
understanding of models, assumptions, and data is piv-
otal [1, 5]. O’Dwyer et al. [21] have demonstrated a
sustainable energy management system to manage the
flow of data between individual machine learning mod-
els, districts, and cities; however a general and scalable
solution for the construction of cross-domain models re-
mains unrealised, impeding the ability to reproduce re-
sults as well as adapt and combine existing models [18].
The World Avatar (TWA) project [16] creates an
ecosystem that enables the transparent integration of
heterogeneous models and data, thereby improving
interoperability between various formats and soft-
ware [22]. It leverages technologies from the Semantic
Web stack to create a distributed dynamic knowledge
graph, which by design is well suited to effectively ad-
dress cross-domain questions. Ontologies provide un-
ambiguous definitions of concepts and relationships to
describe relevant data and computational agents. These
agents act as executable knowledge components which
render the graph inherently dynamic. They share a
common worldview to ensure self-consistency and ac-
complish tasks such as updating the graph, simulat-
ing systems, or transmitting responses to the physical
world. Agents can represent black box, grey box or
physics-based models and can also wrap around ex-
isting software or third-party application programming
interface (APIs) to make them available semantically.
With an initial focus on chemical and process engi-
neering [14, 23, 24], TWA has evolved into a versa-
tile tool to address decarbonisation questions in the en-
ergy sector [25, 26, 22, 27], overcome cross-domain in-
teroperability challenges in smart cities and city plan-
ning [28, 29, 30, 31], and improve the resilience of
complex systems [32]. Akroyd et al. [16] showed how
a dynamic general-purpose knowledge graph based on
ontologies and autonomous semantic agents is ideally
suited to realising connected digital twins, e.g., to con-
trol real-world assets, perform cross-domain simula-
tions, or conducting geospatial and scenario analyses.
The purpose of this paper is to provide a con-
crete implementation example of how dynamic knowl-
edge graphs can help to realise connected digital twins
by combining previously isolated tools and data. The
World Avatar is used to derive a more holistic energy
perspective for smart cities by connecting (1) a knowl-
edge graph-native receding horizon control for the heat
generation of a district heating network with (2) emis-
sion dispersion simulations to understand the impact of
various heat generation strategies on air pollution and
(3) detailed building energy modelling, by making the
City Energy Analyst available semantically.
The structure of this paper is as follows: Section 2
provides an overview of the current energy modelling
landscape for cities, together with its challenges, and
an introduction to The World Avatar dynamic knowl-
edge graph. Section 3 develops new ontologies and soft-
ware capabilities to address identified interoperability
gaps using TWA. Section 4 highlights the results from
the connected digital twin implementation and section 5
concludes the work.
2. Background
Holistic smart city energy modelling encompasses a
broad spectrum of considerations, from energy forecast-
ing to generation optimisation as well as the assessment
of potential consequences of proposed scenarios. This
section provides an overview of previous research and
the status quo in each of these fields. Each topic is
introduced independently, following conventional com-
munity practices; however, as depicted in Fig. 1, these
2
Figure 1: Towards holistic smart city energy modelling. Conventional approaches address relevant aspects like building energy, operations control,
or air pollutant dispersion separately. Individual analyses, simulations, or optimisation often remain isolated, disregarding interdependencies and
overlaps in their inputs and/or outputs. The World Avatar dynamic knowledge graph connects related data and computational agents, enabling
unified and automated modelling based on a consistent world view. While input agents assimilate real world data into the graph, update agents act
upon instantiated information.
silos are resolved in the following sections using a dy-
namic knowledge graph approach. Semantic Web tech-
nologies and The World Avatar, which enable this inte-
gration, are also introduced.
2.1. Energy system modelling for smart cities
Energy systems are increasingly intertwined and de-
mand a comprehensive approach to drive overall re-
source efficiency and decrease emissions [4, 6]. They
are closely tied to numerous key challenges of the
twenty-first century, including security, affordability,
and resilience of energy supply as well as socio-
economic and environmental concerns, ranging from lo-
cal air and water pollution to global sustainability initia-
tives [7, 33]. A holistic energy system approach is im-
perative to provide an efficient and reliable low-carbon
energy supply, comprising the planning and scheduling
of diverse energy carriers, such as electricity, gas, heat-
ing, and cooling [12].
In this context, established methods to model en-
ergy systems are being challenged by several emerging
themes, as extensively discussed in the literature [17, 3,
2, 33]: increased sector-coupling and interactions be-
tween energy vectors at various scales (e.g., from multi-
national, national, community scale down to building
level); rising flexibility of demand driven by new tech-
nologies such as smart meters and load shifting; en-
hanced integration of intermittent renewable resources,
with the resulting need for more temporal detail; dis-
tributed generation and an increasing share of pro-
sumers, with the resulting need for higher spatial gran-
ularity. With increasing connectivity, the ‘internet of
energy’ has emerged as overarching paradigm to elim-
inate waste in the power system [34], requiring mod-
elling frameworks to optimise across scales, considering
multiple spatial and temporal resolutions [7, 3]. Numer-
ous studies have shown that artificial intelligence can
help to improve grid performance and optimise energy
distribution and control in integrated systems [35]; how-
ever, interpretability of models and digital twins is also
gaining importance [36]. A review of recent smart grid
and digital twin efforts is provided by Sifat et al. [37].
A strong push towards open simulation and planning
tools has been observed in recent years as vital building
blocks for transparent modelling approaches that bridge
scales and domains [12, 2]. While it has been demon-
strated that open-source modelling frameworks and data
platforms are often on par with proprietary or commer-
cial models [3], impediments to interoperability persist
due to technical and market barriers. Diverse require-
3
ments and limitations (e.g., regional scope) of individual
tools, coupled with variations in applicability to specific
problems and scenarios, pose a real challenge in inte-
grating data and models [12, 38]. To address these chal-
lenges, semantic approaches have been proposed, such
as by Li and Hong [38], who developed a framework for
grid-interactive efficient buildings which are responsive
to grid pricing or carbon signals to achieve energy and
carbon neutrality.
2.2. The City Energy Analyst
The City Energy Analyst (CEA) is an established
open-source computational framework for urban en-
ergy system analysis [39], offering insights into build-
ings’ overall energy demand, heating and cooling re-
quirements, etc. as well as on-site renewable energy
generation potentials. It has a global user base and
has been applied to numerous case studies across the
world [40, 41, 42, 43, 44].
The CEA toolkit comes with built-in databases, con-
taining several assumptions required to run simulations.
The databases include information about building prop-
erties (i.e., OpenStreetMap (OSM) [45] building foot-
print, height, and building usage) as well as environ-
mental data such as weather and terrain [39]. Although
this approach enables users to conduct simulations with-
out the requirement for specific input data, the built-in
assumptions may not always be representative. Priori-
tising broad applicability over the integration of actual
building-specific characteristics is a deliberate design
choice, inherent to many top-down energy assessment
tools.
2.3. Interoperability gaps
Interoperability is the ability of different systems, de-
vices, or applications to exchange and use information
effectively and collectively. While technical interoper-
ability within the energy sector is quite well-established,
there is a need for more informational, functional, and
business interoperability [20]. Fragmented platforms
dominate the energy modelling landscape, with no uni-
fied approach to harmonise data models or knowledge
across all domains of the value chain [46]. This poses
challenges for data integration, model validation, sce-
nario comparisons, policy evaluation, and often re-
sults in biased or subpar overall system performance,
as decision-makers lack valuable information to assess
certain cross-domain trade-offs or co-benefits of dif-
ferent scenarios. Just to name two examples, cross-
domain interoperability would allow studying the ef-
fects of extreme weather events, such as heat waves,
floods, storms, or earthquakes, on both the built envi-
ronment and smart grid infrastructure, aiding in iden-
tifying potential weak points and enhancing systemic
resilience [32]. Secondly, emission analyses could ex-
tend beyond the established assessment of overall green
house gas emission amounts to explore detailed disper-
sion patterns of individual air pollutants as the result of
different energy provision strategies, by incorporating
location and weather data.
Current interoperability gaps can be addressed by
adopting standards and frameworks to facilitate com-
munication and collaboration among different mod-
elling tools, stakeholders, and platforms or enhancing
information exchange using common data models, on-
tologies, and Semantic Web technologies [5]. While the
first approach remains focused on the broader energy
domain (e.g., by incorporating solar panels, battery stor-
age, heat pumps, boilers and electric vehicles) [46], the
latter one is in principle capable to connect seamlessly
with any related domain, such as transport, agriculture
and industrial production [47].
2.4. Dispersion modelling
Dispersion models can broadly be categorised as box
models, Gaussian plume models, or advanced physical
models [48, 49, 50, 51]. Given their popularity and abil-
ity to incorporate a wide variety of input data [48, 49],
e.g., complex terrains and buildings in the dispersion
pathway, a Gaussian plume model is selected for this
work, assuming that pollutant concentrations follow a
Gaussian distribution. Specifically, AERMOD [52], a
steady-state Gaussian plume model, also deployed by
the United States Environmental Protection Agency to
assess air pollution, is chosen due to available source
code, good documentation, the support for multiple
emission sources, and achievable input data require-
ments (i.e., to ensure the availability of all required in-
puts to run the model).
AERMOD has been applied and validated for a wide
variety of conditions: flat and complex terrains [53,
54, 55], various time scales [56] and emission sources,
such as a cement complex [57] or a coal-fired power
plant [58], and many more available in the literature.
To account for the effect of buildings on the disper-
sion of air pollutants, AERMOD incorporates a vali-
dated downwash model to capture relevant turbulence
effects [59].
2.5. The World Avatar dynamic knowledge graph
As introduced by Akroyd et al. [16], The World
Avatar aims to create a digital ‘avatar’ of the world.
4
This vision of an all-encompassing world model is cur-
rently worked towards using Semantic Web technology,
following a general-purpose dynamic knowledge graph
(dKG) approach [8].
The Semantic Web [60] is an extension of the World
Wide Web with the aim of creating an interoperable
‘web of data’, making web content machine-readable
by adding structured metadata. It builds on the use
of ontologies and the Resource Description Frame-
work (RDF) [61] for representing such metadata. An
ontology provides an explicit description of a specific
domain by formally defining relevant concepts and re-
lationships between them. Due to strict formalisation,
ontologies support unambiguous data sharing and reuse,
and enable reasoning and inference of implicit infor-
mation. Representing data using ontologies results in
the formation of directed graphs, known as knowledge
graphs (KGs), where nodes define concepts, instances,
or data, and edges denote their relationships. KGs pro-
vide extensible data structures well suited to represent
arbitrarily structured data. Using Internationalised Re-
source Identifiers (IRIs), KG resources can be uniquely
identified, allowing data to be distributed across the
web, while maintaining unambiguous links between en-
tities. Such Linked Data [62, 63] supports FAIR data
principles [15] and enhances the discoverability of in-
formation. Knowledge graphs can be stored in triple
stores, such as Blazegraph [64], which are designed to
host RDF data in the form of subject-predicate-object
triples. SPARQL [65] is a query language designed to
interact with semantic information and can be used to
query and update these stores.
Beyond the capabilities of conventional KGs, such as
DBpedia or Wikidata, TWA also includes semantically
annotated computational capabilities, so-called agents,
which operate upon instantiated entities and make the
graph inherently dynamic [23]. Computational agents
within TWA can be seen as executable knowledge com-
ponents and perform diverse tasks, such as assimilating
real-world data, performing calculations, updating the
graph, or transmitting responses to the physical world.
A derived information framework (DIF) [66] provides
a KG-native solution to track data dependencies and
manage information flow within TWA. Offering granu-
lar data provenance on an instance level, it provides de-
tails about the origin of any information and the agent
responsible for its acquisition. By representing intrin-
sic dependencies within the KG, the DIF enables au-
tonomous data handling, allowing information to cas-
cade across the graph.
The combination of ontological descriptions, instan-
tiated data, and autonomous agents makes TWA a pow-
erful, extensible, and FAIR-compliant system for rep-
resenting and reasoning about complex domains of
knowledge. As everything is connected, the design sup-
ports an interoperable ecosystem of connected digital
twins (i.e., tools and services) to describe the behaviour
of interconnected systems of systems. TWA is modular
and scalable by design, supporting both decentralisation
and interoperability across heterogeneous data sources
and software.
2.6. Existing ontologies
Within TWA, ontologies serve as modular compo-
nents to represent and connect knowledge and data from
different domains. Existing ontologies are reused where
applicable, and new ontologies are proposed for identi-
fied gaps. This approach honours existing domain ex-
pertise and ensures compatibility with established com-
munity understanding, while satisfying requirements of
the provided data and target use case.
After reviewing the literature, it became evident that
relevant ontologies for the target use case are either not
publicly available or do not adequately address the re-
quired level of detail in the domain of interest: Al-
though numerous ontologies have been proposed to rep-
resent temporal concepts and/or time dependent mea-
surements [67, 68, 69, 70], no efficient representation
for large amounts of time series data could be identified.
For utility network applications, available ontologies
can capture detailed 3D topography, topology, and func-
tional properties [71, 72, 73]; however, no conceptuali-
sation for dynamic operations data to support the opti-
mal coordination of district energy resources is publicly
available [74, 75]. Ontology-based efforts to represent
air pollution dispersion data exist [14, 23, 76], but lack
compatibility with GeoSPARQL [77] for geospatial fea-
tures, hindering their applicability to new locations; fur-
thermore, a semantic solution for storing pollution con-
centrations, often provided in raster format by mod-
elling software, has not been devised [48, 49, 50, 51], a
gap we aim to fill in this study. GeoSPARQL forms the
de-facto standard for representing and querying geospa-
tial data on the Semantic Web and provides an extension
to the SPARQL query language for processing geospa-
tial data. A more detailed review of previous ontology
efforts is provided in supplementary material SM.2.
3. Implementation
In this section we devise new domain ontologies and
computational agents to operationalise real-world data
related to the optimal control of a district heating sys-
tem as well as the atmospheric dispersion of associated
5
air pollutants. The proposed agents exchange data and
interact with one another via instantiated ontologies in
the dKG, promoting a shared understanding among all
actors. In contrast to conventional API-based methods,
the explicit semantics of ontologically defined commu-
nication establish a uniform, public, and well-controlled
framework for expressing and interpreting data. This
eliminates ambiguity in data representation, allowing
agents to navigate and comprehend information consis-
tently.
3.1. Ontologies
Four new ontologies are proposed to represent time
series data as well as relevant aspects of district heating
network operations, emission dispersion and the build-
ing energy domain. We adopt a bottom-up approach,
with a primary focus on representing real operations
data from our industrial partner, Stadtwerke Pirmasens,
and capturing outputs from the City Energy Analyst.
However, a sufficient level of generality is maintained to
ensure reusability beyond the target use case. Reusing
established ontologies maximises the benefits of pre-
viously conceptualised domain knowledge and linked
data, and requires the use of predefined concept and
relationship names. Furthermore, typical naming con-
ventions suggest the use of singular terms for concept
names, relationships to be prefixed with ‘has’, camel
case for composite phrases, among other field specific
ones.
The consistency of all proposed ontologies has been
verified using the HermiT reasoner [78]. For a formal
representation of the ontologies using description log-
ics [79], please refer to the supplementary material. The
codified [80] versions are publicly accessible on GitHub
in OWL format: https://github.com/cambridge-
cares/TheWorldAvatar/tree/main/JPS_Ontolog
y/ontology.
3.1.1. Time series ontology
This work elaborates an initial approach for a light-
weight time series ontology [81], primarily to include a
description of forecasts and how they have been derived.
The key structure of the ontology is provided in Fig. 2.
While SAREF’s extension [82] acknowledges that
any entity (e.g., measurement) could be represented as
either single value or time series, our approach consid-
ers cases where entities also have associated time series
forecasts. Hence, the domain of hasTimeSeries and
hasForecast remains unconstrained. The Forecast
concept is the central entity to represent any forecast and
is associated with both a TimeSeries concept holding
the actual predicted values and further meta data about
the forecast to ensure proper provenance information.
This includes the ForecastingModel used to derive
the prediction, the length of the historical time series
used for training and/or scaling, and the forecast hori-
zon. Concepts from OM and the W3C time ontology
are used to represent corresponding units and tempo-
ral entities, such as the interval of the forecast horizon.
The ForecastingModel concept captures key aspects
of how forecasts are calculated, including the used train-
ing TimeSeries to fit the model, potential covariates
to be used when creating a forecast, and whether the
data should be scaled when creating predictions (i.e.,
as required by many neural methods). Previously fit-
ted models can be incorporated by specifying resolv-
able URLs for both the saved model and checkpoint
files (e.g., pickled pytorch models). Otherwise, default
forecast models can be specified using a certain label.
Further details on how the forecasting agent uses the
ontology are provided in section 3.2.1 below. In con-
trast to the SAREF extension [82], no explicit restric-
tion on the frequency of the represented data is imposed;
however, a Frequency concept is included, along with
a resampleData property, indicating whether a time
series needs to be resampled when creating a forecast
(e.g., to comply with frequency requirements of certain
forecasting techniques).
3.1.2. District heating network ontology
This ontology aims to conceptualise district heating
network operations and has been designed based on op-
erational practices from Stadtwerke Pirmasens. While
previous works often focus on static topology and 3D
representations [72, 73], the shared information does
not contain detailed geo-references to describe the grid
structure (e.g., pipes, connectors). Moreover, the op-
erations of the grid are rather dynamic, including cus-
tomers’ heat demand and flow temperature profiles. As
the work by Li et al. [75] is not publicly available for di-
rect reuse, an ontology is proposed to capture essential
aspects of district heating operations, leveraging some
design choices from previous works.
The three key concepts in the ontology
are HeatingNetwork, HeatProvider, and
HeatGenerator. The HeatingNetwork connects
HeatProvider with Consumer instances to satisfy
their HeatDemand (i.e., instantiated as time series).
While the location of an individual Consumer presently
remains undisclosed and is, hence, not explicitly
modelled, any HeatProvider is connected to the grid
via a GridConnection with observable properties.
These properties include pressure as well as flow and
6
https://github.com/cambridge-cares/TheWorldAvatar/tree/main/JPS_Ontology/ontology
https://github.com/cambridge-cares/TheWorldAvatar/tree/main/JPS_Ontology/ontology
https://github.com/cambridge-cares/TheWorldAvatar/tree/main/JPS_Ontology/ontology
ts:Forecast
ts:hasForecast
ts:hasForecastingModel
ts:hasInputTimeInterval
time:Interval
ts:hasOutputTimeInterval
ts:hasCovariate
owl:Thing
ts:TimeSeries
ts:hasTimeSeries
xsd:string xsd:string
ts:hasRDB ts:hasTimeUnit
ts:has
TimeSeries
ts:Forecasting
Model
ts:hasModelURL
rdfs:label
om:Unit
om:
hasUnit
ts:has
Training
TimeSeries
ts:hasCheckpointURL
ts:scaleData
time:Duration
ts:Frequency
rdfs:subClassOf
ts:resampleData
xsd:boolean
xsd:string
xsd:boolean
xsd:string
xsd:string
Concept
Object property Data property
Literal
New / Re- used concept or relationshipColour coding:
Figure 2: Time series ontology. The OntoTimeSeries ontology provides a light-weight representation for time series data within TWA. It further
includes a general description of related forecasting concepts. Depicted labels denote existing/proposed concept and relationship names, with all
referenced namespaces being declared in Appendix A.
return om:Temperature, and provide insights into the
temperature spread at these locations and, consequently,
corresponding feed-in heat amounts.
While the HeatProvider concept is kept gen-
eral and represents any entity supplying heat to
the grid, two subclasses relevant to the given use
case are defined, namely MunicipalUtility and
IncinerationPlant. The geospatial location of a
HeatProvider is not explicitly modelled as part of this
ontology. Instead, geo-references are established via
links to corresponding building instances with detailed
geometrical and geospatial information as part of the
derivation markup as described in section 3.3 and illus-
trated in Fig. 3. This approach keeps the OntoHeatNet-
work ontology completely free of geospatial informa-
tion, by reusing the capabilities of existing ontologies.
An IncinerationPlant provides a certain
ProvidedHeatAmount to the grid based on a sup-
ply contract with the grid operator (for details
refer to Fig. SM.1 in the supplementary mate-
rial). A MunicipalUtility company can own
multiple HeatGenerators, including conventional
HeatBoilers and combined heat and power (CHP)
GasTurbines. Each HeatGenerator is associated
with a GeneratedHeatAmount and a corresponding
ConsumedGasAmount concept, according to the gen-
erator’s efficiency and used ocp:FuelType. Relevant
costs are represented on both an individual generator
and operator level. As CO2 emissions directly influence
operating expenses (OPEX) due to emission certificate
cost, they are modelled explicitly as part of OntoHeat-
Network, while other air pollutants are conceptualised
as part of OntoDispersion. Similarly, the electricity
co-generation of a GasTurbine is captured, since
respective revenue offsets heat generation OPEX. A
detailed overview of the hierarchical cost structure
and its components is provided in Fig. SM.2 in the
supplementary material. An Availability concept is
introduced to account for periods of plant shut-downs
or required idle times of individual generators. Most
properties will be instantiated as time series to ac-
count for dynamic conditions (i.e., fluctuating prices,
time-dependent heat demand) and to align with the
hourly-resolved optimisation strategy applied by the
municipal utility operator in the target use case.
3.1.3. Dispersion ontology
This light-weight ontology aims to provide seman-
tic markup for dispersion simulation data to create
machine-readable inputs and outputs for agents and
to foster cross-domain interoperability between various
models. The key concepts of OntoDispersion are lo-
cated in the bottom of Fig. 3, including their intended
link to the OntoHeatNetwork ontology.
A geospatial Scope concept specifies the simulation
domain for the dispersion calculation and is defined as a
subclass of geo:Feature to enable various geospatial
querying and processing capabilities via GeoSPARQL.
By using concepts from GeoSPARQL, the ontology is
designed to be as robust as possible and easily extend-
7
OntoDispersion
oh:Grid
Connection
OntoCAPE:hasPart
oh:HeatProvider
oh:hasUpstream
GridConnection
oh:hasDownstream
GridConnection
oh:Municipal
Utility
oh:Incineration
Plant
oh:Heat
Generator
OntoCAPE:
isOwnerOf
oh:HeatBoiler
oh:GasTurbine
om:Pressure
OntoCAPE:
Thermodynamic
StateProperty
oh:has
ObservableProperty
oh:Availability
oh:hasOperating
Availability
oh:Generated
HeatAmount
oh:Provided
HeatAmount
oh:hasProvided
HeatAmount
oh:HeatDemand
oh:has
Heat
Demand
oh:suppliesHeatTo
ocp:hasFuelType
ocp:FuelType
oh:CO2Factor
oh:has
CO2Factor
oh:CostIn
TimeInterval
oh:has
Consumed
GasAmount
oh:CoGen
ElectricityAmount
oh:hasCoGen
ElectricityAmount
oh:CoGenRevenue
InTimeInterval
oh:hasOperating
Availability
OntoCAPE:
hasRevenue
oh:provides
HeatTo
oh:has
Generated
HeatAmount
om:Temperature
oh:Consumed
GasAmount
oh:FuelUnitCost
oh:hasUnitPrice
oh:operates
Concept Literal
New / Re- used concept or relationshipColour coding:
oh:Consumer
oh:Heating
Network
To be instantiated according to om or ts ontology
disp:Static
PointSource
disp:Emission
disp:emits
Building IRI
disp:has
OntoCityGML
CityObject
disp:PollutantID
disp:has
PollutantID
disp:NOx
disp:PM2.5
disp:PM10
disp:Dispersion
Output
disp:has
PollutantID disp:z
disp:
hasHeight
disp:Dispersion
Raster
disp:hasDispersionRaster
om:Height
disp:Scope geo:Feature
Object property Data property rdfs:subClassOfInstance
OntoCAPE:
hasCost
Figure 3: Coupled heating network and dispersion modelling ontology. The OntoHeatNetwork ontology conceptualises the operations of a district
heating network, while OntoDispersion provides a semantic description of air pollutant emission dispersion. Both ontologies can be linked through
shared building instances to support geospatial emission analyses of district heating operations. Depicted labels denote existing/proposed concept
and relationship names, with all referenced namespaces being declared in Appendix A.
able to different areas. Versatile geospatial capabilities
are essential to query which StaticPointSources,
emitting one or more pollutant types, are located within
a certain scope of interest. StaticPointSources link
to corresponding building instances, which describe the
actual geometries of the emission outlets. Each in-
stance of DispersionOutput holds information on
a set of raster data (DispersionRaster) for any ar-
bitrary combination of pollutant type (PollutantID)
and simulated height (z). In this work, we do not at-
tempt to materialise raster data as RDF triples. Instead,
any DispersionRaster instance simply provides the
metadata (e.g., name of a GeoTIFF file) of a raster
stored in an associated PostGIS database. Thus, an
agent querying for dispersion raster data would obtain
the metadata via a SPARQL query and perform a subse-
quent SQL query to obtain the underlying raster values.
3.1.4. Building energy ontology
Energy considerations are important in city master
planning, and buildings are the main user of urban en-
ergy. There are ontologies for urban building energy
modelling [83, 84, 85] and master planning [28], but
none really links the two fields. This ontology aims to
bridge this gap to facilitate information exchange be-
tween these closely related domains.
An extract of the proposed OntoUBEMMP on-
tology is shown in Fig. 4. The key concepts
are dabgeo:Building, EnergyConsumption and
EnergySupply, while the core building concept is
shared with the OntoBuiltEnv ontology to facilitate
interoperability between this energy-specific perspec-
8
dabgeo:Building
bs:WallFacade
bs:RoofFacade
ub:SolarDevice
ub:EnergySupply
ub:Electricity
Supply
ub:HeatSupply
ub:producesEnergy
bs:hasFacade
ub:Energy
Consumption
ub:Heating
Consumption
ub:Electricity
Consumption
ub:Cooling
Consumption
om:Energy
ub:consumes
Energy
om:Areaub:hasSolar
SuitableArea
ub:hasTheo-
reticalEnergy
Production
Concept New / Re- used concept or relationshipColour coding:
To be instantiated according to om or ts ontology
Object property rdfs:subClassOf
ub:SolarCollectorub:PVPanel ub:PVTCollector
Figure 4: Urban Building Energy Modelling and Master Planning ontology. The OntoUBEMMP ontology represents key concepts in the nexus of
urban energy modelling and master planning, including building energy demands and solar potentials. Depicted labels denote existing/proposed
concept and relationship names, with all referenced namespaces being declared in Appendix A.
tive and a more comprehensive building description
provided by OntoBuiltEnv. An EnergyConsumption
concept is linked to its applicable building instance
via a consumesEnergy relationship to represent a
building’s energy demand. Renewable energy sources
should be taken into consideration during master plan-
ning to help offset EnergyConsumption. For exam-
ple, a building can be equipped with SolarDevices
on its bs:RoofFacade and its bs:WallFacade. For
buildings with suitable areas for solar generation (i.e.,
hasSolarSuitableArea links to a non-zero area),
the hasTheoreticalEnergyProduction relationship
connects relevant areas with their potential Energy-
Supply via the respective SolarDevice. There
are different subclasses of EnergySupply, namely
ElectricitySupply and HeatSupply, depending on
the type of SolarDevice that could be installed. Instal-
lation of PVPanel will generate ElectricitySupply,
whereas SolarCollector will generate HeatSupply
and the hybrid PVTCollector will generate both. The
ontology uses OntoTimeSeries to instantiate Energy-
Consumption and EnergySupply concepts as well as
their subclasses to account for variable demand patterns
or changing weather conditions.
3.2. Agents
Several agents have been developed. An overview
of all involved agents is provided in Fig. 5 and de-
scribed below. All agents are packaged as individ-
ual Docker services to foster distributed and platform-
agnostic deployment (e.g., remotely in the cloud, as im-
plemented for this use case). Compared to rather mono-
lithic modelling approaches, the proposed design based
on chained atomic agents allows for emerging complex-
ity within system of systems architectures, without be-
ing constrained to strictly linear dependencies.
We were given actual historical operations data for
a municipal district heating network of a midsize Ger-
man town, Pirmasens. Based on this data, the district
heating grid is instantiated per OntoHeatNetwork, us-
ing 2020 historical time series data. Utilising the in-
stantiated time series data, a forecasting agent can be
used to predict any quantity, including the community’s
HeatDemand. A district heating optimisation agent
then generates a cost-optimised generator dispatch strat-
egy to satisfy the forecast HeatDemand, considering
both internal heat generators and sourcing from an ex-
ternal waste incineration plant. The respective amounts
of burned natural gas as well as heat from waste in-
cineration are then converted by an emission estimation
agent into corresponding NOx, PM2.5, and PM10 emis-
sion streams. Together with the associated location in-
formation, these emission streams form inputs to an dis-
persion modelling agent to create a steady-state emis-
sion dispersion map using actual historical wind data.
All agents are implemented as derivation agents based
on the DIF [66] and communicate directly via the dKG
to ensure unambiguous provenance tracking of how a
9
loop
optional
optional
The Word
Avatar KGUser
Opti-
misation
Trigger Agent
AERMOD
Agent
DIF
Fore-
casting
Agent
Emission
Agent
Opti-
misation
Agent
HTTP request with
1) Simulation start time
2) Optimisation horizon
3) Number of timesteps
Update pure input instances to make existing derivations out-of-date
(forecast, heat generation optimisation, emission and dispersion)
KG updated
Request
update for
dispersion
derivation
Check whether dispersion derivation still up-to-date
Invoke agent
to update
derivation
[for each
timestep]
Request
update for
emission
derivation
Check whether emission derivation still up-to-date
Invoke agent to update
forecast derivation Update outputs (heat demand and grid temperatures)
KG updated
Forecast derivation updated
Invoke agent to update heat
generation optimisation derivation Update provided and generated heat
as well as consumed gas amounts
KG updated
Heat generation optimisation derivation updated
Invoke agent to update emission estimation derivation Update emission
instance
KG updated
Emission derivation updated
Emission
derivation
up-to-date Update dispersion output
KG updatedDispersion
derivation
updated
Dispersion
derivation
up-to-date
Optimisation run
complete
[if dispersion
derivation
out-of-date]
Forecast
derivation is
out-of-date,
causing heat
generation
optimisation and
emission
derivations to be
both out-of-date
[if emission
derivation
out-of-date]
Figure 5: Agent interplay. Sequence diagram of all agents involved in the heat generation optimisation with integrated emission dispersion
modelling (depicted for case of already instantiated derivation markups).
certain output has been derived and which inputs it de-
pends on. We introduce an optimisation trigger agent to
coordinate between a user and the automated forecast-
ing, optimisation, and subsequent emission dispersion
simulation.
While the dynamic load forecasting and supply-side
optimisation use actual historical data, the City Energy
Analyst agent can provide general insights into the en-
ergy performance of buildings in case historical data
is not available: utilising building-specific construction
characteristics and weather data, various energy demand
and generation profiles can be estimated. This com-
plementary perspective provides valuable insights into
building-resolved heat demands, e.g., relevant to anal-
yse any potential extension of the district heating grid.
3.2.1. Forecasting agent
This agent provides generic forecasting capabilities
as part of TWA: it can retrieve instantiated time series,
predict future values, and instantiate respective fore-
casts back into the dKG using the OntoTimeSeries on-
tology. Based on the Python library Darts [86], the
agent supports forecasting via a wide range of meth-
ods, ranging from classical white box models (e.g., es-
tablished statistical methods such as autoregressive inte-
grated moving-average (ARIMA) models and its deriva-
10
tives) to black box machine learning techniques (e.g.,
state-of-the-art transformer models), as well as grey box
approaches such as Facebook’s Prophet [87].
The required input to derive any forecast comprise
the instance associated with the time series to pre-
dict, a ts:ForecastingModel describing the predic-
tion model to use, the target ts:Frequency of the fore-
cast to be created, a time:Interval denoting the tar-
get forecast horizon, and a time:Duration denoting
the historical data length to use for fitting and/or scal-
ing of the historical time series data prior to creating the
forecast. New forecasts are instantiated with relevant
metadata, such as input and output time intervals as well
as potentially applicable unit, as depicted in Fig. 2.
The agent supports most forecasting models offered
by Darts. The model to use is determined by the
instantiated mark-up of the corresponding ts:Fore-
castingModel instance in the dKG, with Prophet be-
ing the default for arbitrary time series. Predictions
with and without covariates are supported, depend-
ing on whether ts:hasCovariate relationships are
present for the target model instance. Additionally,
custom models can be trained and stored within TWA
for future forecasting. This involves creating custom
ts:ForecastingModel instances with specific prop-
erties, such as resolvable URLs for saved model files,
relevant covariates, and scaling parameters. Thus, the
agent offers both out-of-the-box forecasting capabilities
and the flexibility to leverage custom fine-tuned models
as needed.
The agent can predict any arbitrary time series and
is used to forecast heat demand and grid temperatures
of a district heating network, using temporal fusion
transformers (TFT) [88] in the context of the current
work. Compared to many other deep learning methods,
attention-based TFTs provide better explainability and
interpretability thanks to insights into underlying atten-
tion weights, which indicate what a model focuses on.
The fitted TFT models are more accurate than the fine-
tuned SARIMAX models deployed previously [89], es-
pecially for longer forecast horizons. Further imple-
mentation details as well as a detailed forecast perfor-
mance comparison are provided in supplementary ma-
terial SM.5.1.
3.2.2. District heating optimisation agent
This agent leverages a previously developed optimi-
sation routine to minimise total heat generation cost for
a district heating provider [89]. The optimisation fol-
lows a hierarchical approach based on merit-order prin-
ciple to determine the OPEX-optimised short-term heat
generation mix for a system comprised of multiple gas
boilers, a CHP gas turbine as well as external heat sourc-
ing from a waste incineration plant. This study show-
cases the capability of dKGs in enabling connected dig-
ital twins to derive more comprehensive energy perspec-
tives and semantically integrates the existing model into
TWA. The interested reader is referred to Hofmeister
et al. [89], where the effectiveness of the optimisation
has been demonstrated based on real-world operations
data.
While the initial optimisation relied on internally
created SARIMAX predictions for key inputs, this
agent increases modularity and fosters a micro-service
architecture enabled through task-oriented connected
digital twins by using externally instantiated fore-
casts. The agent requires five ts:Forecast and one
time:Interval instance to perform an optimisation.
The interval specifies the optimisation horizon, describ-
ing the period for which to derive the optimal dispatch
strategy, while the five forecasts denote the forecasted
oh:HeatDemand and four om:Temperatures (i.e., rep-
resenting flow and return temperatures at the waste in-
cineration and municipal heating plant) over this period,
respectively. Besides these key inputs, further infor-
mation are queried from the dKG during agent opera-
tion. Upon successful optimisation, the following re-
sults are instantiated back into the dKG according to the
OntoHeatNet ontology: a oh:ProvidedHeatAmount
instance describing the heat amount to be sourced
from the waste incineration plant; an oh:Generated-
HeatAmount and oh:ConsumedGasAmount instance
for each gas boiler and CHP gas turbine denoting the
heat amount to be provided and corresponding gas
amount to be consumed by each heat generator, respec-
tively; an oh:CoGenElectricityAmount instance de-
scribing the amount of co-generated electricity by the
gas turbine while providing the required amount of heat;
an oh:Availability instance for each heat provider
indicating its anticipated availability in the coming
time steps. All optimisation outputs are instantiated
as ts:Forecast instances for the respective concepts
to not interfere with instantiated actual historical data.
Newly created optimisation outputs automatically over-
write previously instantiated ones.
Upon first invocation of the agent, historical gas con-
sumption, heat generation, and (if applicable) electric-
ity generation data is queried to fit data-driven genera-
tor specific gas consumption and co-generation models
to be used during the optimisation. These models will
be reused for all subsequent optimisation requests. The
results of the agent implementation have been verified
against previous optimisation results [89].
11
3.2.3. Emission estimation agent
This agent estimates the emission rates associated
with heat production from burning natural gas (i.e., in
gas boilers or the CHP gas turbine) or waste (i.e., in the
waste incineration plant). For the time being, the assess-
ment is limited to NOx, PM2.5, and PM10 as the major
airborne emissions [90, 91, 92] and relies on literature-
based emission factors instead of detailed combustion
models for this proof-of-concept and due to the absence
of detailed information on the waste incineration plant
internals.
Implemented as a derivation agent, all required in-
puts need to be available in the KG. This includes
one dh:ProvidedHeatAmount or, alternatively, one or
more dh:ConsumedGasAmounts representing the (op-
timised) time series for externally sourced heat or con-
sumed gas amounts, respectively. A collection of con-
sumed gas amounts resembles multiple gas boilers and
gas turbines housed within the same building, emit-
ting exhausts through a shared chimney. A disp:Sim-
ulationTime marks the timestamp for which to esti-
mate the emissions, i.e., for which later to simulate the
emission dispersion. Lastly, a disp:StaticPoint-
Source instance specifies the location at which the es-
timated emissions will be emitted.
During assessment, the time series values for pro-
vided heat or consumed gas corresponding to the tar-
get disp:SimulationTime are extracted. If multiple
dh:ConsumedGasAmounts are given, their individual
values are added together and processed collectively.
Subsequently, emission factors are applied to convert
the energy amounts into corresponding mass flow rates
for NOx, PM2.5, and PM10, as required for the air pol-
lutant dispersion simulation. The flue gas stream is
treated as hot air, using typical values from the litera-
ture. Please refer to supplementary material SM.5.2 for
more details about the estimation methods. All outputs
are instantiated according to the OntoDispersion ontol-
ogy as disp:Emission instances with associated quan-
tities for mass flow rate of pollutant as well as temper-
ature and density of the exhaust stream. One emission
instance per pollutant type is created.
3.2.4. Dispersion modelling agent
This agent utilises AERMOD [52, 53] to simulate
the dispersion of various air pollutants in a specific
area of interest. It considers instantiated wind and
emission stream data (i.e., mass flow rate, tempera-
ture) from multiple point sources to generate emission
concentration maps. Upon invocation, the agent per-
forms three key steps: querying relevant inputs from
the KG, executing AERMOD using this information,
and finally instantiating the results back into the KG.
As shown in Fig. 6, the key inputs are disp:Scope and
disp:SimulationTime. The disp:Scope defines the
polygon of the simulation domain (i.e., a rectangle) and
disp:SimulationTime determines the time step of in-
terest for which to run the dispersion calculation. Note
that this input is shared with the emission estimation
agent.
The agent requires at least one instance of
disp:StaticPointSource (e.g., a chimney emitting
pollutants) that is located within disp:Scope to simu-
late a plume. Instances of disp:StaticPointSource
are not linked directly to dispersion derivations in or-
der to facilitate future use cases involving mobile point
sources (e.g., ships), which may move in and out of
disp:Scope, making explicit markups very difficult
to maintain. Instead, the agent uses disp:Scope to
obtain relevant buildings and emission sources within
the simulation area for the relevant timestamp. The
disp:SimulationTime is also used to query the ac-
tual (historical) weather data for that given time. Hav-
ing retrieved all necessary information, the agent com-
poses relevant input files and executes AERMOD. Sub-
sequently, the agent processes the dispersion results into
raster form and updates the disp:DispersionOutput
time series instance in the dKG.
Although we use AERMOD in this work, it could
be swapped with any other dispersion model (e.g.,
EPISODE [93]) with minimal changes to the overall
workflow outlined in Fig. 6. It is inevitable that a new
agent would need to be developed; however, the pro-
posed ontology would still suffice to represent relevant
concepts (e.g., disp:DispersionRaster).
3.2.5. City Energy Analyst agent
This agent calculates various aspects of a building’s
energy performance using CEA as its simulation en-
gine. To overcome limitations with built-in CEA as-
sumptions, actual building stock data from the dKG
are incorporated to allow for building-specific analy-
ses, namely a building’s geometry and usage, the ge-
ometry of surrounding buildings, weather, and terrain
data. The implementation maintains CEA’s broad ap-
plicability while adopting a building-resolved bottom-
up approach.
Upon invocation, the CEA agent attempts to retrieve
the geometry of the target building(s) (i.e., specified
by the building IRI(s) in the received HTTP request)
as well as the geometries of the surrounding build-
ings. Retrieved geometries from TWA replace CEA’s
default OSM footprints. Subsequently, the agent will at-
tempt to retrieve building specific usage data from TWA
12
to produce most meaningful energy consumption pro-
files. After retrieving building level input data, the agent
attempts to retrieve actual weather information at the
target location from the dKG. Available local weather
data supersedes CEA’s default behaviour of interpolat-
ing weather information based on a few selected loca-
tions within its own database. Lastly, the agent attempts
to retrieve terrain data, specifically, the elevation of the
land surrounding the target building(s) from TWA, to
replace CEA’s default terrain input of a fixed elevation.
Only a building’s geometry is strictly necessary for the
agent to run successfully. In cases where other inputs
cannot be retrieved from TWA (i.e., surroundings, us-
age, weather, terrain), it proceeds to run CEA with its
corresponding default assumptions. For details please
refer to supplementary material SM.5.3.
After running the simulations, relevant results are in-
stantiated according to OntoUBEMMP in TWA. The
various building energy demands (i.e., heating, cool-
ing, electricity, grid) are instantiated as ub:Energy-
Consumption instances. The agent also provides so-
lar potential estimates for various types of solar gen-
erators: PV panels, flat plate and evacuated tube solar
collectors, and combined flat plate and evacuated tube
PV-thermal collectors. These generators are instantiated
as the corresponding subclasses of ub:SolarDevice,
with their associated energy potentials instantiated as
ub:EnergySupply entities. The suitable area for in-
stalling solar devices is instantiated via the ub:has-
SolarSuitableArea property.
3.3. Linked agents for cross-domain interoperability
Individual agents are chained together via their input
and output instances using TWA’s native derived infor-
mation framework. This ensures that whenever a spe-
cific piece of information is requested from the dKG,
all dependent upstream inputs are scrutinised first to de-
termine if they are still up-to-date or require updating
before retrieval. We leverage this infrastructure to au-
tomatically simulate associated air pollution dispersion
whenever a new heat generation optimisation is com-
puted and corresponding emission streams get instanti-
ated.
As illustrated in Fig. 5, the optimisation trigger agent
acts as external input agent. To initiate an optimisation
run, an HTTP POST request is expected, specifying 1)
the optimisation start time, 2) the optimisation horizon
(i.e., the number of time steps to be considered within
each optimisation), and 3) the number of subsequent
time steps to optimise in total. Upon receiving and veri-
fying a request, the agent creates/updates corresponding
instances within the dKG, and an update is requested
from the dispersion modelling agent (also referred to
as AERMOD agent due to the implemented model) via
the DIF. The DIF then assesses whether an up-to-date
dispersion instance already exists by comparing the in-
stantiation timestamp of the derivation instance against
the ones of corresponding inputs. If necessary, an up-
date is requested, in which case the DIF works back-
wards through the dependencies: dispersion simulations
depend on emission estimation outputs, which depend
on the heat generation optimisation, itself dependent
on heat demand and grid temperature forecasts. The
DIF initiates updates by invoking the associated agents,
starting with the forecasting agent, responsible for the
most upstream derivation, to ensure a proper cascade
of all dependent information. Once all information is
up-to-date, the initially requested dispersion outputs are
simulated, which marks the end of the current optimisa-
tion run. This loop is repeated until the number of time
steps to optimise is reached.
To ensure automated information cascading, deriva-
tion markups need to be instantiated at the instance
level, as illustrated in Fig. 6. It is important to note that
this figure is a simplified representation for readability,
with a more detailed diagram provided in Fig. SM.3 in
the supplementary material. The optimisation trigger
agent instantiates initial inputs for time:Interval,
time:Duration, and ts:Frequency for each re-
quested optimisation run (and updates them accordingly
for subsequent time steps). Additionally, the agent pro-
grammatically creates the depicted derivation markup if
not already present and requests an initial assessment
from responsible agents to generate corresponding out-
puts for further markup.
One derivation instance is created for each re-
quired forecast, i.e., one oh:HeatDemand and four
om:Temperature instances denoting the flow and re-
turn temperatures at the municipal heat and waste in-
cineration plant, respectively. The derivation outputs
(i.e., updated ts:Forecast instances) are then collec-
tively marked up as inputs to the heat generation op-
timisation derivation. Given the accuracy of the fine-
tuned TFT models, significant over- or underpredic-
tions of required heat supply are improbable. Nonethe-
less, the ‘network storage effect’ could offset such rare
occurrences by utilising the network as short-term en-
ergy storage. After optimising the generation dispatch
based on the provided forecasts, multiple outputs are
instantiated by the district heating optimisation agent,
including one oh:ProvidedHeatAmount and several
oh:ConsumedGasAmount instances, representing the
time series of external heat provision from the waste in-
cineration plant and gas consumption of several internal
13
time:Duration
time:Interval
Forecasting
Agent
Forecast
derivation
:isDerived
From:isDerived
Using
ts:Forecast
(grid temperature)
ts:Forecast
(heat demand)
:belongsTo
Heat
generation
optimisation
derivation
:isDerivedFrom
DH
Optimisation
Agent
:isDerived
Using
oh:Provided
HeatAmount
:belongsTo
:isDerivedFrom
:belongsTo
Emission
Agent
:isDerived
Using
Dispersion
derivation
disp:Dispersion
OutputAERMOD
Agent
:isDerived
Using
ts:Frequency
:isDerivedFrom
om:Temperature
oh:HeatDemand
ts:Forecasting
Model
oh:Consumed
GasAmount
oh:Generated
HeatAmount
Emission
derivation
disp:Emission
either or (i.e., one
derivation per forecast quantity)
Agent
Pure Input
Instance
Instances and
relationships
required by
Derived Information
Framework
Output
Instance
Legend
:belongs
To
disp:SimulationTime
:isDerived
From
Request
update
and query om:MassFlow disp:Scope
:isDerived
From
disp:StaticPointSource
disp:
emits
:isDerived
From
om:has
Quantity
Figure 6: Derivation chain (simplified). Schematic depiction of knowledge graph native instance markup to resemble a model predictive control
loop, coupled with automated air pollutant dispersion modelling. All referenced namespaces are declared in Appendix A, with not explicitly stated
prefixes referring to OntoDerivation.
heat generators, respectively.
Subsequently, two individual emission derivations
are marked up to account for different emission fac-
tors used for waste and natural gas burning when es-
timating associated emission rates. Different deriva-
tion instances also account for different locations of the
respective pollutant streams, as each derivation is de-
rived from a disp:StaticPointSource, which intro-
duces a geospatial reference to the dynamic optimisa-
tion. The estimated disp:Emission outputs are used
as source terms by AERMOD to simulate pollutant dis-
persion maps. Although not explicitly marked up as
14
inputs, the agent requires at least one StaticPoint-
Source (e.g., a chimney) within the disp:Scope of in-
terest. The disp:SimulationTime instance represents
the time for which to simulate the emission dispersion,
and matches the first time step of the forecast and opti-
mised heat generation.
The district heating optimisation agent is designed
to handle each optimisation request independently, i.e.,
without consideration of any preceding requests. The
sole exception to this behaviour occurs when two con-
secutive requests are an hour apart. In this scenario,
the second request is treated as dependent on the pre-
vious one, facilitating the tracking of relevant system
state variables, such as cumulative profit from ongoing
gas turbine activity. This enables a dKG-native receding
horizon optimisation implementation, representing the
first model predictive control-style application within
TWA. While demonstrated for energy dispatch with in-
tegrated emission modelling, similar derivation chains
can automate various other (cross-domain) smart city
workflows. Although the current implementation relies
on an optimisation trigger agent as external input agent,
this can easily be replaced with autonomous agents in
the future.
4. Comprehensive energy perspective
This section describes results and insights from the
connected digital twin. Utilising the developed ontolo-
gies and semantic agents, connected via the derived in-
formation framework, allows for novel insights and ca-
pabilities, such as 1) the knowledge graph-native con-
trol of a district heating system, 2) the refinement of
building energy analyses with latest instantiated build-
ing stock data, and 3) cross-domain insights into heat
generation induced air pollutants dispersion.
The World Avatar offers a versatile visualisation in-
terface to explore and interact with the underlying data,
and supports both Mapbox (i.e., mainly for geographic
information) and Cesium (i.e., mainly for detailed geo-
metrical representations) as well-established visualisa-
tion frameworks; however, it is crucial to understand
that the presented visualisations are not the digital twin
itself. Instead, the digital twins are a dynamic collec-
tion of knowledge, data, and models embedded in the
dynamic knowledge graph running in the background,
with the visualisation being only one way to access it.
Further options include a mobile app [94], virtual reality
goggles, and a question answering system [95] besides
a unified SPARQL endpoint.
The integrated visualisation interface provides both
map-based and (real-time) dashboard features. While
map-based visualisations can help to understand the
geospatial distribution of energy demand or the implica-
tions of certain heat sourcing strategies on air pollution,
dashboards focus on time series data and offer more de-
tails about the current operational state of assets, such
as the latest historical and forecast heat demand or the
optimised generation strategy to satisfy it.
4.1. Resource-efficient heat provision
The municipal district heating network has been in-
stantiated based on actual data, including historical op-
eration, weather, and market conditions. Operations
data include details about the grid itself as well as at-
tached heat providers, while market conditions cover
electricity spot, gas, or CO2 certificate price time se-
ries. Instantiated heating network data comprises the to-
tal heat demand profile of all attached customers, opera-
tional boundaries of the grid (e.g., minimum volumetric
flow rate to ensure hydrodynamic stability), and connec-
tion properties for the municipal heating and waste in-
cineration plant, such as observed flow and return tem-
peratures. Plant data includes information about the
buildings hosting individual heat generators, along with
their design characteristics (e.g., rated thermal power)
and dynamic properties, such as time series of gener-
ated heat and electricity as well as consumed gas. The
integration across scales (i.e., from city level to de-
tailed boiler specifications) as well as the inherent dy-
namism due to the dKG-native control implementation
combines and exceeds the capabilities of isolated energy
system modelling and geographic information system-
based approaches.
Figure 7 presents a snapshot of the dynamic heat de-
mand forecast dashboard. It shows the recent municipal
heat demand history of all district heating consumers as
well as the latest 24-hour demand forecast. The dash-
board updates automatically with each new forecast, of-
fering real-time insights into the latest operational state.
In addition to time series visualisation (right), a gauge
indicates the current state relative to operational/ob-
served minimum and maximum values (left).
Figure 8 illustrates the optimal dispatch of three con-
ventional heat boilers, one CHP gas turbine, and exter-
nal heat sourcing from the nearby waste incineration
plant to satisfy the predicted demand (refer to Fig. 7).
Currently, the demand of approximately 10.5 MW h is
met through external sourcing and one heat boiler, while
the remaining heat generators remain idle. Based on the
projected demand as well as anticipated electricity spot
prices, the CHP gas turbine is expected to be the main
contributor to heat production as of in 4 hours, with mi-
15
Figure 7: Heat demand forecast. Dashboard view of the latest historical and forecast heat demand at any given time step. The historical load profile
is shown left of the dashed line, with predicted values to its right.
Figure 8: Optimised heat generation. Dashboard view of the cost-optimised heat distribution across generators and external sources, considering a
waste incineration plant, three conventional gas boilers, and a gas turbine (based on forecast heat demand).
nor support from the waste incineration plant and one
additional gas boiler.
4.2. City energy analyses and scenario planning
As actual (historical) energy data are not always
available, an alternative approach is needed to estimate
relevant quantities and gain insights on a broader scale,
such as city level. The CEA agent provides estimates
for various aspects of buildings’ energy performance,
such as demands for different types of energy and on-
site solar generation potentials. Compared to the of-
ficial CEA toolkit, actual building stock (i.e., building
geometry, geometry of surrounding buildings, property
usage) as well as weather and terrain data are used for
the underlying simulations (where available) to derive
building-specific estimates. The outputs of the agent are
16
Figure 9: Visualisation of annual heating demand of each building simulated by the CEA agent. While map-based visualisation allows for quick
identification of buildings with high/low heating demand, time series support the inspection of load profiles for individual buildings.
instantiated and attached to the corresponding building
in TWA and can be inspected via its unified visualisa-
tion interface. Figure 9, for example, shows the an-
nual heating demand for a selected neighbourhood in
Pirmasens, allowing for a quick identification of build-
ings (and areas) with high/low heating demand. Fur-
ther simulation results, such as photovoltaic potential or
gross-floor area specific values are provided in supple-
mentary material SM.6.
While Fig. 9 offers rather qualitative insights, the
credibility of the results has been evaluated for both
electricity consumption and on-site solar PV potential.
The assessment compares instantiated agent results with
actual historical consumption data or the official PV
potential estimates provided by the state of Rhineland-
Palatinate [96], respectively. By leveraging more gran-
ular building an weather information from TWA, signif-
icant accuracy improvements compared to native CEA
(i.e., the unaltered CEA toolkit) can be achieved. The
mean absolute percentage error (MAPE) relative to the
above benchmarks could be reduced from 57.6% to
13.7% and from 28.1% to 12.9% for annual electricity
consumption and solar PV potential, respectively. This
improvement outlines the value of our bottom-up ap-
proach to remove default assumptions in the underly-
ing CEA toolkit where actual data is available from the
dKG.
Beyond cumulative annual figures, the CEA agent
also provides both overall heat demand and solar poten-
tial time series, which facilitate the assessment of pos-
Figure 10: Heating load profiles simulated by the CEA agent. Heating
demand and solar generation time series can be used to evaluate po-
tential energy savings of on-site solar collector installations (depicted
for a typical day in March).
sible energy savings achievable with the installation of
solar collectors. A basic analysis could explore utilising
heat from rooftop solar panels to directly offset a build-
ing’s heating demand, without factoring in any thermal
storage. This simplistic assessment provides a prelimi-
nary estimate for remaining heating demand from alter-
native sources such as gas or district heating, together
with the potential energy savings conferred by on-site
generation (see Fig. 10). This capability can help to de-
velop highly-granular heat maps of a city’s heating de-
17
(a) Heat generation related NOx emission dispersion as of 09 Dec 2020 06:00 UTC.
(b) Heat generation related NOx emission dispersion as of 09 Dec 2020 07:00 UTC.
Figure 11: Integrated emission dispersion simulation. The integrated simulation of heat generation induced air pollutants provides insights into air
pollution implications of various heat generation/sourcing strategies, considering actual (historical) weather data.
18
mand (with or without considering on-site generation of
solar energy), e.g., as currently required for the munic-
ipal heat planning initiative in Germany. The building-
resolved insights exceed the accuracy of most publicly
available dataset, which are usually restricted to sim-
ple raster maps with 50 × 50 m or 100 × 100 m reso-
lution [97]. Combined with actual district heating grid
location data, this information can be used to evaluate
potential grid extension scenarios, both with regards to
the total geospatially distributed heat demand as well as
prevalent heat demand profiles considering actual build-
ing usage patterns.
4.3. Impact on air quality
Beyond insights into the energetic behaviour of build-
ings and their optimised heat provision, a key strength
of The World Avatar lies in generating cross-domain in-
sights: Emission dispersion simulations are triggered
automatically by each heat generation optimisation to
immediately understand potential impacts of the pro-
jected heat sourcing strategy, comprising multiple lo-
cations, on the exposure of various parts of the sur-
rounding population to associated airborne emissions.
This proof-of-concept predominantly focuses on con-
necting the dynamic cost optimisation with geospatial
emission implications and a detailed investigation of po-
tential health consequences remains yet unexplored.
To mirror actual operating conditions, the dynamic
optimisation is deployed with an hourly resolution.
Hence, also the emission dispersion is simulated for
each optimised hour, producing one instantiated disper-
sion raster per air pollutant and elevation of interest.
As generic Gaussian plume model, AERMOD supports
various emission types; however, this work focuses on
NOx, PM2.5, and PM10 as major pollutants (see sec-
tion SM.5.2 in the supplementary material for details),
with NOx typically exhibiting the highest proportional
concentrations. While the dispersion at arbitrary ele-
vations relative to the underlying terrain can be stud-
ied, our focus centres on ground level (i.e., 0 m of el-
evation), given its importance for pedestrians and the
general public. A summary of relevant parameters for
the AERMOD simulations is provided in supplemen-
tary material SM.1.
Instantiated dispersion maps can be overlaid with
buildings or population density data to inspect various
aspects and potential implications of heat generation in-
duced emissions. In Fig. 11, this capability is show-
cased, displaying NOx emission values in conjunction
with instantiated building stock, where the colours indi-
cate the usage of properties, with blue representing pre-
dominantly residential buildings. The figure illustrates
a historical scenario across two consecutive hours, dur-
ing which the start-up of the CHP gas turbine has been
deemed profitable by the optimisation routine.
Figure 11(a) illustrates a heat provision situation
where the majority of heat is sourced from the waste
incineration plant situated in the North of the town. Fig-
ure 11(b) depicts the heat generation one hour later, in-
cluding the active gas turbine located at the municipal
heating plant in the Southern part of the town. Given
the different geo-locations of various heat sources, dis-
tinct exposure scenarios emerge based on the chosen
heat provision strategy due to the incorporation of wind
data. Despite similar wind conditions and comparable
maximum concentrations, both situations exhibit signif-
icantly different exposure potentials. In the first sce-
nario, multiple residential buildings face relatively high
NOx concentrations, whereas these areas are shifted to
regions without residential buildings in the second sce-
nario.
Figure 12 depicts the exposure of the town’s popu-
lation to additional air pollution for two different heat
sourcing strategies by overlaying the dispersion visual-
isation over the population density raster: Despite sim-
ilar weather conditions, certain geographic areas expe-
rience significantly different exposure. The strategy de-
picted in Fig. 12(a) sources most of the heat from the
waste incineration plant located at the outskirts, result-
ing in relatively higher but more remote emissions. The
strategy shown in Fig. 12(b) distributes heat sourcing
more evenly between the two available sites, resulting
in lower overall concentrations; however, the municipal
heating plant’s plume affects central areas with higher
population density. This trade-off shall be addressed
in the future by coupling heat generation with poten-
tial health implications for the surrounding population
within a multi-objective optimisation framework.
Beyond the sole map view, virtual sensors can be
placed at arbitrary locations to study air pollution expo-
sure over time. These sensors extract data from underly-
ing raster files and display corresponding values as time
series for the respective pollutant types in the visualisa-
tion side panel, as illustrated in Fig. 13. The depicted
emission profiles for different pollutants look similar
due to deploying a Gaussian dispersion model. Results
could not explicitly be validated against actual local air
quality readings due to a lack of available historical data
(i.e., sensor readings) within the area of interest. How-
ever, given the numerous previous calibration studies of
AERMOD, we believe that the derived values possess
at least indicative meaning. Moreover, simulated val-
ues align well with applicable emission thresholds (see
Table SM.5 in the supplementary material) as well as
19
(a) NOx emission dispersion as of 10 Dec 2020 04:00 UTC (b) NOx emission dispersion as of 10 Dec 2020 17:00 UTC
Figure 12: Emission exposure. The integrated dispersion simulation provides insights into the exposure of certain parts of the population to
additional air pollutants as a result of heat generation. Illustrated for simulated optimisation results overlaid with the population density raster.
Figure 13: Emission time series. Virtual sensors allow inspecting simulated emission concentrations for arbitrary locations of interest as well as
time scales.
published hourly and daily mean readings reported by
the waste incineration plant operator [98]. While the
current dispersion model is intentionally simplistic for
this proof-of-concept, the workflow can easily accom-
20
modate more sophisticated models in future iterations
without significant modifications.
5. Conclusions
In this work, we demonstrate the capabilities of The
World Avatar dynamic knowledge graph to create com-
prehensive energy perspectives for smart cities, thereby
bridging domains in the nexus of energy, the built envi-
ronment, and atmospheric dispersion. We connect de-
tailed energy analyses for individual buildings with dy-
namic control of a municipal district heating system and
simulation of associated emissions, all seamlessly inte-
grated within one interoperable semantic system.
We have extended the ontological coverage of The
World Avatar with knowledge models to describe time
series and forecasts, district heating network opera-
tions, building energy characteristics, and air pollutant
dispersion and leverage these ontologies to instantiate
real-world data. We have developed multiple semantic
agents to act upon the instantiated data and deploy them
in a connected fashion to provide the proof-of-concept
for the first model predictive control application within
TWA. We have implemented a generic forecasting agent
and use it as part of this knowledge graph-native reced-
ing horizon optimisation to minimise the total heat gen-
eration cost of a district heating network. The outputs
of the optimisation are directly linked with an integrated
emission dispersion model to understand the impact of
various heat generation and sourcing strategies on air
pollution. The showcased degree of interoperability and
automation, spanning from energy forecasting to cost-
optimal generator dispatch to airborne emission disper-
sion, is enabled by the automated tracking of dependen-
cies between various agent and data instances within the
dynamic knowledge graph.
Furthermore, the City Energy Analyst is made avail-
able as part of TWA to provide valuable information
about buildings’ energy demands and on-site genera-
tion potentials with regards to solar energy. The devel-
oped agent offers a flexible enhancement to the orig-
inal CEA toolkit by utilising latest instantiated build-
ing and weather data from the knowledge graph to re-
duce the dependency on default assumptions; thus, pro-
moting a data-driven bottom-up approach for compre-
hensive energy assessments of building stock at various
levels (e.g., building, district, and city level). This per-
spective complements the dynamic generation optimisa-
tion with a more strategic angle, relevant to analyse any
potential expansion of the district heating grid to drive
low-carbon heating solutions.
This work shows that semantically linked agents of-
fer great potential to resemble the behaviour of com-
plex systems, address interoperability challenges holis-
tically, and implement automatable cross-domain smart
city workflows, even beyond the energy sector. This
study primarily focuses on showcasing an implementa-
tion example of this capability, placing less emphasis on
quantitative analyses. By integrating real-world sensor
data, the accuracy of deployed models, such as AER-
MOD or the City Energy Analyst, can continuously be
refined and, reversely, virtual sensors can easily be de-
ployed to fill gaps in the actual sensor landscape with
simulated readings, thereby creating a truly interoper-
able and dynamic cyber-physical system of connected
digital twins.
Acknowledgements
This research was supported by the National Re-
search Foundation, Prime Minister’s Office, Singapore
under its Campus for Research Excellence and Techno-
logical Enterprise (CREATE) programme. Part of this
work was also supported by Towards Turing 2.0 un-
der the EPSRC Grant EP/W037211/1. M. Hofmeister
acknowledges financial support provided by the Cam-
bridge Trust and CMCL. M. Kraft gratefully acknowl-
edges the support of the Alexander von Humboldt Foun-
dation.
The authors express gratitude to the Stadt Pirmasens,
especially mayor Michael Maas and his team, as well
as the Stadtwerke Pirmasens, with Christoph Dörr and
his team, for their invaluable collaboration and generous
support in sharing relevant data, enhancing the depth
and quality of this research. This work also leverages
data from©GeoBasis-DE/LVermGeoRP 2023. Further-
more, the authors express gratitude to L.F. Ding and
G.H. Xiao for their valuable contributions, particularly
in sharing the Ontop mapping and engaging in helpful
discussions.
The graphical abstract leverages material designed by
macrovector/Freepik. For the purpose of open access,
the author has applied a Creative Commons Attribution
(CC BY) licence to any Author Accepted Manuscript
version arising.
Nomenclature
AERMOD AMS/EPA regulatory model (air dispersion
model)
API Application programming interface
ARIMA Autoregressive integrated moving average
21
CEA City Energy Analyst
CHP Combined heat and power
DIF Derived information framework
dKG Dynamic knowledge graph
EEA European environment agency
GeoSPARQL Geographic query language for RDF
data
IRI Internationalized resource identifier
KG Knowledge graph
LSTM Long short-term memory
MAPE Mean absolute percentage error
ME Maximum error
NO2 Nitrogen dioxide
NOx Nitrogen oxides
OM Ontology of units of measure
OPEX Operating expense
OSM OpenStreetMap
PM10 Particulate matter less than 10 µm in diameter
PM2.5 Particulate matter less than 2.5µm in diameter
PM Particulate matter
PV Photovoltaics
RDF Resource description framework
RMSE Root mean square error
SAREF Smart Applications REFerence ontology
SARIMAX Seasonal autoregressive integrated moving
average with exogenous regressors
SMAPE Symmetric mean absolute percentage error
SPARQL SPARQL protocol and RDF query language
SQL Structured query language
TFT Temporal fusion transformer
TWA The World Avatar (dynamic knowledge graph)
W3C World Wide Web Consortium
WHO World Health Organization
Declaration of Generative AI and AI-assisted tech-
nologies in the writing process
During the preparation of this work the authors used
ChatGPT in order to enhance the readability and lan-
guage of the manuscript. After using this tool, the au-
thors reviewed and edited the content as needed and take
full responsibility for the content of the publication.
Data and code availability
All the codes developed are available on The World
Avatar GitHub repository:
https://github.com/cambridge-cares/TheWorldAvatar.
Developed ontologies can be found in the ontology sub-
directory and instructions to reproduce the use case are
detailed in the Pirmasens repository.
Conflicts of interest
There are no conflicts of interest to declare.
Appendix A. Namespaces
deriv:
disp:
oh:
ts:
ocp:
OntoCAPE:
OntoPowSys:
ub:
bs:
contract:
dabgeo:
geo:
om:
owl:
rdf:
rdfs:
time:
xsd:
References
[1] O. Inderwildi, C. Zhang, X. Wang, M. Kraft, The impact of
intelligent cyber-physical systems on the decarbonization of en-
ergy, Energy & Environmental Science 13 (2020) 744–771.
doi:10.1039/c9ee01919g.
[2] L. Hofbauer, W. McDowall, S. Pye, Challenges and opportu-
nities for energy system modelling to foster multi-level gover-
nance of energy transitions, Renewable and Sustainable Energy
Reviews 161 (2022) 112330. doi:10.1016/j.rser.2022.112330.
[3] M. Fodstad, P. Crespo del Granado, L. Hellemo, B. R. Knudsen,
P. Pisciella, A. Silvast, C. Bordin, S. Schmidt, J. Straus, Next
frontiers in energy system modelling: A review on challenges
and the state of the art, Renewable and Sustainable Energy Re-
views 160 (2022) 112246. doi:10.1016/j.rser.2022.112246.
[4] A. Berjawi, S. Walker, C. Patsios, S. Hosseini, An evaluation
framework for future integrated energy systems: A whole en-
ergy systems approach, Renewable and Sustainable Energy Re-
views 145 (2021) 111163. doi:10.1016/j.rser.2021.111163.
22
https://github.com/cambridge-cares/TheWorldAvatar
https://github.com/cambridge-cares/TheWorldAvatar/tree/main/JPS_Ontology/ontology
https://github.com/cambridge-cares/TheWorldAvatar/tree/main/JPS_Ontology/ontology
https://github.com/cambridge-cares/TheWorldAvatar/tree/main/Deploy/stacks/Pirmasens
https://www.theworldavatar.com/kg/ontoderivation/
https://www.theworldavatar.com/kg/ontoderivation/
https://www.theworldavatar.com/kg/ontodispersion/
https://www.theworldavatar.com/kg/ontodispersion/
https://www.theworldavatar.com/kg/ontoheatnetwork/
https://www.theworldavatar.com/kg/ontoheatnetwork/
https://www.theworldavatar.com/kg/ontotimeseries/
https://www.theworldavatar.com/kg/ontotimeseries/
http://theworldavatar.com/ontology/ontochemplant/OntoChemPlant.owl#
http://theworldavatar.com/ontology/ontochemplant/OntoChemPlant.owl#
http://www.theworldavatar.com/ontology/ontocape/chemical_process_system/CPS_performance/economic_performance.owl#
http://www.theworldavatar.com/ontology/ontocape/chemical_process_system/CPS_performance/economic_performance.owl#
http://www.theworldavatar.com/ontology/ontocape/chemical_process_system/CPS_performance/economic_performance.owl#
http://www.theworldavatar.com/ontology/ontopowsys/PowSysRealization.owl#
http://www.theworldavatar.com/ontology/ontopowsys/PowSysRealization.owl#
https://www.theworldavatar.com/kg/ontoubemmp/
https://www.theworldavatar.com/kg/ontobuildingstructure/
https://www.theworldavatar.com/kg/ontobuildingstructure/
https://spec.edmcouncil.org/fibo/ontology/FND/Agreements/Contracts/
https://spec.edmcouncil.org/fibo/ontology/FND/Agreements/Contracts/
http://www.purl.org/oema/infrastructure/
http://www.opengis.net/ont/geosparql#
http://www.ontology-of-units-of-measure.org/resource/om-2/
http://www.ontology-of-units-of-measure.org/resource/om-2/
http://www.w3.org/2002/07/owl#
http://www.w3.org/1999/02/22-rdf-syntax-ns#
http://www.w3.org/2000/01/rdf-schema#
http://www.w3.org/2006/time#
http://www.w3.org/2001/XMLSchema#
https://doi.org/10.1039/c9ee01919g
https://doi.org/10.1016/j.rser.2022.112330
https://doi.org/10.1016/j.rser.2022.112246
https://doi.org/10.1016/j.rser.2021.111163
[5] V. Reif, T. I. Strasser, J. Jimeno, M. Farre, O. Genest, A. Gy-
rard, M. McGranaghan, G. Lipari, J. Schütz, M. Uslar, S. Vo-
gel, A. Bytyqi, R. Dornmair, A. Corusa, G. Roy, F. Ponci,
A. Dognini, A. Monti, Towards an interoperability roadmap
for the energy transition, e & i Elektrotechnik und Information-
stechnik 140 (2023) 478–487. doi:10.1007/s00502-023-01144-
2.
[6] O. Angelidis, A. Ioannou, D. Friedrich, A. Thomson, G. Fal-
cone, District heating and cooling networks with decentralised
energy substations: Opportunities and barriers for holistic en-
ergy system decarbonisation, Energy 269 (2023) 126740.
doi:10.1016/j.energy.2023.126740.
[7] S. Pfenninger, A. Hawkes, J. Keirstead, Energy systems
modeling for twenty-first century energy challenges, Re-
newable and Sustainable Energy Reviews 33 (2014) 74–86.
doi:10.1016/j.rser.2014.02.003.
[8] M. Q. Lim, X. Wang, O. Inderwildi, M. Kraft, The World Avatar
– A World Model for Facilitating Interoperability, in: O. In-
derwildi, M. Kraft (Eds.), Intelligent Decarbonisation: Can Ar-
tificial Intelligence and Cyber-Physical Systems Help Achieve
Climate Mitigation Targets?, Springer International Publishing,
Cham, 2022, pp. 39–53. doi:10.1007/978-3-030-86215-2_4.
[9] G. F. Huseien, K. W. Shah, A review on 5G technology for smart
energy management and smart buildings in Singapore, Energy
and AI 7 (2022) 100116. doi:10.1016/j.egyai.2021.100116.
[10] A. Tzachor, S. Sabri, C. E. Richards, A. Rajabifard, M. Acuto,
Potential and limitations of digital twins to achieve the Sus-
tainable Development Goals, Nature Sustainability 5 (2022)
822–829. doi:10.1038/s41893-022-00923-7.
[11] B. Wu, W. D. Widanage, S. Yang, X. Liu, Battery digital twins:
Perspectives on the fusion of models, data and artificial intelli-
gence for smart battery management systems, Energy and AI 1
(2020) 100016. doi:10.1016/j.egyai.2020.100016.
[12] N. Wu, F. Zhang, J. Wang, X. Wang, J. Wu, J. Huang, J. Tan,
R. Jing, J. Lin, S. Xie, Y. Zhao, A review on modelling
methods, tools and service of integrated energy systems in
China, Progress in Energy 5 (2023) 032003. doi:10.1088/2516-
1083/acef9e.
[13] H. Y. Quek, F. Sielker, J. Akroyd, A. N. Bhave, A. von
Richthofen, P. Herthogs, C. v. d. L. Yamu, L. Wan, T. Nochta,
G. Burgess, M. Q. Lim, S. Mosbach, M. Kraft, The conundrum
in smart city governance: Interoperability and compatibility in
an ever-growing ecosystem of digital twins, Data & Policy 5
(2023) e6. doi:10.1017/dap.2023.1.
[14] A. Eibeck, M. Q. Lim, M. Kraft, J-Park Simulator: An
ontology-based platform for cross-domain scenarios in pro-
cess industry, Computers & Chemical Engineering 131 (2019)
106586. doi:10.1016/j.compchemeng.2019.106586.
[15] M. D. Wilkinson, M. Dumontier, I. J. Aalbersberg, et al.,
The FAIR Guiding Principles for scientific data manage-
ment and stewardship, Scientific Data 3 (2016) 160018.
doi:10.1038/sdata.2016.18.
[16] J. Akroyd, S. Mosbach, A. Bhave, M. Kraft, Universal Digital
Twin - A Dynamic Knowledge Graph, Data-Centric Engineer-
ing 2 (2021) e14. doi:10.1017/dce.2021.10.
[17] U. ur Rehman, P. Faria, L. Gomes, Z. Vale, Future of en-
ergy management systems in smart cities: A systematic liter-
ature review, Sustainable Cities and Society 96 (2023) 104720.
doi:10.1016/j.scs.2023.104720.
[18] S. Pfenninger, J. DeCarolis, L. Hirth, S. Quoilin, I. Staffell,
The importance of open data and software: Is energy re-
search lagging behind?, Energy Policy 101 (2017) 211–215.
doi:10.1016/j.enpol.2016.11.046.
[19] S. Pfenninger, L. Hirth, I. Schlecht, E. Schmid, F. Wiese,
T. Brown, C. Davis, M. Gidden, H. Heinrichs, C. Heuberger,
S. Hilpert, U. Krien, C. Matke, A. Nebel, R. Morrison,
B. Müller, G. Pleßmann, M. Reeg, J. C. Richstein, A. Shiv-
akumar, I. Staffell, T. Tröndle, C. Wingenbach, Open-
ing the black box of energy modelling: Strategies and
lessons learned, Energy Strategy Reviews 19 (2018) 63–71.
doi:10.1016/j.esr.2017.12.002.
[20] B. Rana, Y. Singh, P. K. Singh, A systematic survey on inter-
net of things: Energy efficiency and interoperability perspec-
tive, Transactions on Emerging Telecommunications Technolo-
gies 32 (2021) e4166. doi:10.1002/ett.4166.
[21] E. O’Dwyer, I. Pan, R. Charlesworth, S. Butler, N. Shah, In-
tegration of an energy management tool and digital twin for
coordination and control of multi-vector smart energy sys-
tems, Sustainable Cities and Society 62 (2020) 102412.
doi:10.1016/j.scs.2020.102412.
[22] T. Savage, J. Akroyd, S. Mosbach, M. Hillman, F. Sielker,
M. Kraft, Universal digital twin – the impact of heat pumps on
social inequality, Advances in Applied Energy 5 (2022) 100079.
doi:10.1016/j.adapen.2021.100079.
[23] X. Zhou, A. Eibeck, M. Q. Lim, N. Krdzavac, M. Kraft,
An agent composition framework for the J-Park Sim-
ulator – a knowledge graph for the process industry,
Computers & Chemical Engineering 130 (2019) 106577.
doi:10.1016/j.compchemeng.2019.106577.
[24] S. D. Rihm, M. K. Kovalev, A. A. Lapkin, J. W. Ager, M. Kraft,
On the role of C4 and C5 products in electrochemical CO2 re-
duction via copper-based catalysts, Energy & Environmental
Science 16 (2023) 1697–1710. doi:10.1039/d2ee03752a.
[25] A. Devanand, G. Karmakar, N. Krdzavac, R. Rigo-Mariani,
E. Y. S. Foo, I. A. Karimi, M. Kraft, OntoPowSys: A
power system ontology for cross domain interactions in an
eco industrial park, Energy and AI 1 (2020) 100008.
doi:10.1016/j.egyai.2020.100008.
[26] T. Savage, J. Akroyd, S. Mosbach, N. Krdzavac, M. Hillman,
M. Kraft, Universal Digital Twin: Integration of national-scale
energy systems and climate data, Data-Centric Engineering 3
(2022) e23. doi:10.1017/dce.2022.22.
[27] J. Atherton, J. Akroyd, F. Farazi, S. Mosbach, M. Q. Lim,
M. Kraft, British wind farm ESS attachments: curtailment re-
duction vs. price arbitrage, Energy & Environmental Science 16
(2023) 4020–4040. doi:10.1039/d3ee01355c.
[28] A. Chadzynski, N. Krdzavac, F. Farazi, M. Q. Lim, S. Li, A. Gri-
siute, P. Herthogs, A. von Richthofen, S. Cairns, M. Kraft,
Semantic 3D City Database — An enabler for a dynamic
geospatial knowledge graph, Energy and AI 6 (2021) 100106.
doi:10.1016/j.egyai.2021.100106.
[29] A. Chadzynski, S. Li, A. Grisiute, F. Farazi, C. Lind-
berg, S. Mosbach, P. Herthogs, M. Kraft, Semantic 3D
City Agents—An intelligent automation for dynamic geospa-
tial knowledge graphs, Energy and AI 8 (2022) 100137.
doi:10.1016/j.egyai.2022.100137.
[30] A. Chadzynski, S. Li, A. Grisiute, J. Chua, M. Hofmeister,
J. Yan, H. Y. Tai, E. Lloyd, Y. K. Tsai, M. Agarwal, J. Akroyd,
P. Herthogs, M. Kraft, Semantic 3D city interfaces—Intelligent
interactions on dynamic geospatial knowledge graphs, Data-
Centric Engineering 4 (2023) e20. doi:10.1017/dce.2023.14.
[31] M. Hofmeister, G. Brownbridge, M. Hillman, S. Mosbach,
J. Akroyd, K. F. Lee, M. Kraft, Cross-domain flood
risk assessment for smart cities using dynamic knowledge
graphs, Sustainable Cities and Society (2023) 105113.
doi:10.1016/j.scs.2023.105113.
[32] Digital Twin Hub, Climate Resilience Demonstrator, 2022.
Available at https://digitaltwinhub.co.uk/credo/c
redo/ (accessed Dec 2023).
[33] J. Wu, J. Yan, H. Jia, N. Hatziargyriou, N. Djilali, H. Sun, In-
23
https://doi.org/10.1007/s00502-023-01144-2
https://doi.org/10.1007/s00502-023-01144-2
https://doi.org/10.1016/j.energy.2023.126740
https://doi.org/10.1016/j.rser.2014.02.003
https://doi.org/10.1007/978-3-030-86215-2_4
https://doi.org/10.1016/j.egyai.2021.100116
https://doi.org/10.1038/s41893-022-00923-7
https://doi.org/10.1016/j.egyai.2020.100016
https://doi.org/10.1088/2516-1083/acef9e
https://doi.org/10.1088/2516-1083/acef9e
https://doi.org/10.1017/dap.2023.1
https://doi.org/10.1016/j.compchemeng.2019.106586
https://doi.org/10.1038/sdata.2016.18
https://doi.org/10.1017/dce.2021.10
https://doi.org/10.1016/j.scs.2023.104720
https://doi.org/10.1016/j.enpol.2016.11.046
https://doi.org/10.1016/j.esr.2017.12.002
https://doi.org/10.1002/ett.4166
https://doi.org/10.1016/j.scs.2020.102412
https://doi.org/10.1016/j.adapen.2021.100079
https://doi.org/10.1016/j.compchemeng.2019.106577
https://doi.org/10.1039/d2ee03752a
https://doi.org/10.1016/j.egyai.2020.100008
https://doi.org/10.1017/dce.2022.22
https://doi.org/10.1039/d3ee01355c
https://doi.org/10.1016/j.egyai.2021.100106
https://doi.org/10.1016/j.egyai.2022.100137
https://doi.org/10.1017/dce.2023.14
https://doi.org/10.1016/j.scs.2023.105113
https://digitaltwinhub.co.uk/credo/credo/
https://digitaltwinhub.co.uk/credo/credo/
tegrated Energy Systems, Applied Energy 167 (2016) 155–157.
doi:10.1016/j.apenergy.2016.02.075.
[34] M. Farhan, T. N. Reza, F. R. Badal, M. R. Islam, S. M.
Muyeen, Z. Tasneem, M. M. Hasan, M. F. Ali, M. H.
Ahamed, S. Abhi, M. M. Islam, S. K. Sarker, S. K. Das,
P. Das, Towards next generation Internet of Energy system:
Framework and trends, Energy and AI 14 (2023) 100306.
doi:10.1016/j.egyai.2023.100306.
[35] Z. Liu, Y. Sun, C. Xing, J. Liu, Y. He, Y. Zhou, G. Zhang,
Artificial intelligence powered large-scale renewable integra-
tions in multi-energy systems for carbon neutrality transition:
Challenges and future perspectives, Energy and AI 10 (2022)
100195. doi:10.1016/j.egyai.2022.100195.
[36] M. Manfren, P. A. James, V. Aragon, L. Tronchin, Lean
and interpretable digital twins for building energy monitoring
– A case study with smart thermostatic radiator valves and gas
absorption heat pumps, Energy and AI 14 (2023) 100304.
doi:10.1016/j.egyai.2023.100304.
[37] M. M. H. Sifat, S. M. Choudhury, S. K. Das, M. H. Ahamed,
S. Muyeen, M. M. Hasan, M. F. Ali, Z. Tasneem, M. M.
Islam, M. R. Islam, M. F. R. Badal, S. H. Abhi, S. K.
Sarker, P. Das, Towards electric digital twin grid: Technol-
ogy and framework review, Energy and AI 11 (2023) 100213.
doi:10.1016/j.egyai.2022.100213.
[38] H. Li, T. Hong, A semantic ontology for representing and quan-
tifying energy flexibility of buildings, Advances in Applied En-
ergy 8 (2022) 100113. doi:10.1016/j.adapen.2022.100113.
[39] J. A. Fonseca, N. Thuy-An, A. Schlueter, F. Marechal, City
Energy Analyst (CEA): Integrated framework for analysis and
optimization of building energy systems in neighborhoods and
city districts, Energy and Buildings 113 (2016) 202–226.
doi:10.1016/j.enbuild.2015.11.055.
[40] M. Mosteiro-Romero, J. A. Fonseca, A. Schlueter, Sea-
sonal effects of input parameters in urban-scale building en-
ergy simulation, Energy Procedia 122 (2017) 433–438.
doi:10.1016/j.egypro.2017.07.459.
[41] G. Happle, J. A. Fonseca, A. Schlueter, Effects of air in-
filtration modeling approaches in urban building energy de-
mand forecasts, Energy Procedia 122 (2017) 283–288.
doi:10.1016/j.egypro.2017.07.323.
[42] P. Mehta, D. Griego, A. Nunez-Jimenez, A. Schlueter, The Im-
pact of self-consumption regulation on individual and commu-
nity solar PV adoption in Switzerland: an agent-based model,
Journal of Physics: Conference Series 1343 (2019) 012143.
doi:10.1088/1742-6596/1343/1/012143.
[43] M. Mosteiro-Romero, D. Maiullari, F. Collins, A. Schlueter,
A. V. Timmeren, District-scale energy demand modeling
and urban microclimate: A case study in The Netherlands,
Journal of Physics: Conference Series 1343 (2019) 012003.
doi:10.1088/1742-6596/1343/1/012003.
[44] G. Happle, Z. Shi, S. Hsieh, B. Ong, J. A. Fonseca, A. Schlueter,
Identifying carbon emission reduction potentials of BIPV in
high-density cities in Southeast Asia, Journal of Physics:
Conference Series 1343 (2019) 012077. doi:10.1088/1742-
6596/1343/1/012077.
[45] OpenStreetMap contributors, Planet OSM dump, 2017. Avail-
able at https://planet.osm.org.
[46] Digital4Grids for the European Commission, Landscape report
on energy and flexibility data models and interoperability across
the sectors of energy, mobility and buildings, 2023. Available at
https://digital-strategy.ec.europa.eu/en/libra
ry/landscape-report-energy-and-flexibility-da
ta-models-and-interoperability-across-sectors-
energy (accessed Jan 2024).
[47] int:net consortium, The Interoperability Network for the Energy
Transition (int:net), 2022. Available at https://intnet.eu/
(accessed Dec 2023).
[48] N. Holmes, L. Morawska, A review of dispersion modelling and
its application to the dispersion of particles: An overview of dif-
ferent dispersion models available, Atmospheric Environment
40 (2006) 5902–5928. doi:10.1016/j.atmosenv.2006.06.003.
[49] M. El-Harbawi, Air quality modelling, simulation, and compu-
tational methods: a review, Environmental Reviews 21 (2013)
149–179. doi:10.1139/er-2012-0056.
[50] S. Khan, Q. Hassan, Review of developments in air qual-
ity modelling and air quality dispersion models, Journal
of Environmental Engineering and Science 16 (2020) 1–10.
doi:10.1680/jenes.20.00004.
[51] G. Oliveri Conti, B. Heibati, I. Kloog, M. Fiore, M. Ferrante,
A review of AirQ Models and their applications for forecasting
the air pollution health outcomes, Environmental Science and
Pollution Research 24 (2017) 6426–6445. doi:10.1007/s11356-
016-8180-1.
[52] A. J. Cimorelli, S. G. Perry, A. Venkatram, J. C. Weil, R. J.
Paine, R. B. Wilson, R. F. Lee, W. D. Peters, R. W. Brode, AER-
MOD: A Dispersion Model for Industrial Source Applications.
Part I: General Model Formulation and Boundary Layer Charac-
terization, Journal of Applied Meteorology 44 (2005) 682–693.
doi:10.1175/jam2227.1.
[53] S. G. Perry, A. J. Cimorelli, R. J. Paine, R. W. Brode, J. C.
Weil, A. Venkatram, R. B. Wilson, R. F. Lee, W. D. Peters,
AERMOD: A Dispersion Model for Industrial Source Appli-
cations. Part II: Model Performance against 17 Field Study
Databases, Journal of Applied Meteorology 44 (2005) 694–708.
doi:10.1175/jam2228.1.
[54] D. J. Carruthers, C. A. McHugh, E. Vanvyve, E. Solazzo, Com-
parison of ADMS and AERMOD meteorological preprocessor
and dispersion algorithms, Proceedings of Air & Waste Man-
agement Association, Air Quality Models: Next Generation of
Models (2009) 28–30.
[55] D. J. Carruthers, M. D. Seaton, C. A. McHugh, X. Sheng,
E. Solazzo, E. Vanvyve, Comparison of the complex terrain
algorithms incorporated into two commonly used local-scale air
pollution dispersion models (adms and aermod) using a hybrid
model, Journal of the Air & Waste Management Association 61
(2011) 1227–1235. doi:10.1080/10473289.2011.609750.
[56] B. Zou, F. Benjamin Zhan, J. Gaines Wilson, Y. Zeng, Per-
formance of AERMOD at different time scales, Simu-
lation Modelling Practice and Theory 18 (2010) 612–623.
doi:10.1016/j.simpat.2010.01.005.
[57] K. Seangkiatiyuth, V. Surapipith, K. Tantrakarnapa, A. W.
Lothongkum, Application of the AERMOD modeling system
for environmental impact assessment of NO2 emissions from a
cement complex, Journal of Environmental Sciences 23 (2011)
931–940. doi:10.1016/S1001-0742(10)60499-8.
[58] M. M. Mokhtar, M. H. Hassim, R. M. Taib, Health risk assess-
ment of emissions from a coal-fired power plant using AER-
MOD modelling, Process Safety and Environmental Protection
92 (2014) 476–485. doi:10.1016/j.psep.2014.05.008.
[59] R. L. Petersen, S. A. Guerra, A. S. Bova, Critical review of
the building downwash algorithms in AERMOD, Journal of
the Air & Waste Management Association 67 (2017) 826–835.
doi:10.1080/10962247.2017.1279088.
[60] T. Berners-Lee, J. Hendler, O. Lassila, The Semantic Web, Sci-
entific American 284 (2001) 28–37.
[61] G. Klyne, J. J. Carroll, Resource Description Framework
(RDF): Concepts and Abstract Syntax. W3C Recommendation
10 February 2004, World Wide Web Consortium (W3C), 2004.
Available at http://www.w3.org/TR/2004/REC-rdf-
concepts-20040210/ (accessed Dec 2023).
24
https://doi.org/10.1016/j.apenergy.2016.02.075
https://doi.org/10.1016/j.egyai.2023.100306
https://doi.org/10.1016/j.egyai.2022.100195
https://doi.org/10.1016/j.egyai.2023.100304
https://doi.org/10.1016/j.egyai.2022.100213
https://doi.org/10.1016/j.adapen.2022.100113
https://doi.org/10.1016/j.enbuild.2015.11.055
https://doi.org/10.1016/j.egypro.2017.07.459
https://doi.org/10.1016/j.egypro.2017.07.323
https://doi.org/10.1088/1742-6596/1343/1/012143
https://doi.org/10.1088/1742-6596/1343/1/012003
https://doi.org/10.1088/1742-6596/1343/1/012077
https://doi.org/10.1088/1742-6596/1343/1/012077
https://planet.osm.org
https://digital-strategy.ec.europa.eu/en/library/landscape-report-energy-and-flexibility-data-models-and-interoperability-across-sectors-energy
https://digital-strategy.ec.europa.eu/en/library/landscape-report-energy-and-flexibility-data-models-and-interoperability-across-sectors-energy
https://digital-strategy.ec.europa.eu/en/library/landscape-report-energy-and-flexibility-data-models-and-interoperability-across-sectors-energy
https://digital-strategy.ec.europa.eu/en/library/landscape-report-energy-and-flexibility-data-models-and-interoperability-across-sectors-energy
https://intnet.eu/
https://doi.org/10.1016/j.atmosenv.2006.06.003
https://doi.org/10.1139/er-2012-0056
https://doi.org/10.1680/jenes.20.00004
https://doi.org/10.1007/s11356-016-8180-1
https://doi.org/10.1007/s11356-016-8180-1
https://doi.org/10.1175/jam2227.1
https://doi.org/10.1175/jam2228.1
https://doi.org/10.1080/10473289.2011.609750
https://doi.org/10.1016/j.simpat.2010.01.005
https://doi.org/10.1016/S1001-0742(10)60499-8
https://doi.org/10.1016/j.psep.2014.05.008
https://doi.org/10.1080/10962247.2017.1279088
http://www.w3.org/TR/2004/REC-rdf-concepts-20040210/
http://www.w3.org/TR/2004/REC-rdf-concepts-20040210/
[62] T. Berners-Lee, Linked data - design issues, 2006. Available at
http://www.w3.org/DesignIssues/LinkedData.html
(accessed Dec 2023).
[63] C. Bizer, T. Heath, T. Berners-Lee, Linked data: The story so
far, in: A. Sheth (Ed.), Semantic Services, Interoperability and
Web Applications: Emerging Concepts, IGI Global, 2011, pp.
205–227. doi:10.4018/978-1-60960-593-3.ch008.
[64] Blazegraph, 2020. Available at https://blazegraph.com
(accessed Dec 2023).
[65] C. B. Aranda, O. Corby, S. Das, L. Feigenbaum, P. Gearon,
B. Glimm, S. Harris, S. Hawke, I. Herman, N. Humfrey,
N. Michaelis, C. Ogbuji, M. Perry, A. Passant, A. Polleres,
E. Prud’hommeaux, A. Seaborne, G. T. Williams, SPARQL
1.1 Overview, W3C Recommendation 21 March 2013, World
Wide Web Consortium (W3C), 2013. Available at https:
//www.w3.org/TR/sparql11-overview/ (accessed Dec
2023).
[66] J. Bai, K. F. Lee, M. Hofmeister, S. Mosbach, J. Akroyd,
M. Kraft, A derived information framework for a dy-
namic knowledge graph and its application to smart cities,
Future Generation Computer Systems 152 (2024) 112–126.
doi:10.1016/j.future.2023.10.008.
[67] S. Cox, C. Little, J. R. Hobbs, F. Pan, Time Ontology in OWL,
2020. Available at https://www.w3.org/TR/owl-time/
(accessed Dec 2023).
[68] World Wide Web Consortium, Semantic Sensor Network On-
tology, 2017. Available at https://www.w3.org/TR/vocab-
ssn/ (accessed Dec 2023).
[69] K. Janowicz, A. Haller, S. J. Cox, D. Le Phuoc, M. Lefrançois,
SOSA: A lightweight ontology for sensors, observations, sam-
ples, and actuators, Journal of Web Semantics 56 (2019) 1–10.
doi:10.1016/j.websem.2018.06.003.
[70] ETSI, SAREF: the Smart Applications REFerence ontology,
2020. Available at https://saref.etsi.org/core/ (ac-
cessed Dec 2023).
[71] T. Becker, C. Nagel, T. H. Kolbe, Semantic 3D Model-
ing of Multi-Utility Networks in Cities for Analysis and 3D
Visualization, Springer Berlin Heidelberg, 2012, p. 41–62.
doi:10.1007/978-3-642-29793-9_3.
[72] X. Xu, H. Cai, Domain Ontology for Utility Infrastructure: Cou-
pling the Semantics of CityGML Utility Network ADE and Do-
main Glossaries, Journal of Computing in Civil Engineering 35
(2021) 04021011. doi:10.1061/(asce)cp.1943-5487.0000977.
[73] T. El-Diraby, H. Osman, A domain ontology for con-
struction concepts in urban infrastructure products,
Automation in Construction 20 (2011) 1120–1132.
doi:10.1016/j.autcon.2011.04.014.
[74] J.-L. Hippolyte, Y. Rezgui, H. Li, B. Jayan, S. Howell,
Ontology-driven development of web services to support dis-
trict energy applications, Automation in Construction 86 (2018)
210–225. doi:10.1016/j.autcon.2017.10.004.
[75] Y. Li, Y. Rezgui, S. Kubicki, An intelligent semantic sys-
tem for real-time demand response management of a ther-
mal grid, Sustainable Cities and Society 52 (2020) 101857.
doi:10.1016/j.scs.2019.101857.
[76] C. Metral, G. Falquet, K. Karatzas, Ontologies for the integra-
tion of air quality models and 3D city models, arXiv preprint
arXiv:1201.6511, 2012. doi:10.48550/arXiv.1201.6511.
[77] OGC, OGC GeoSPARQL - A Geographic Query Language for
RDF Data, 2012. Available at https://www.ogc.org/stan
dards/geosparql/ (accessed Dec 2023).
[78] Data & Knowledge Group, HermiT OWL Reasoner, 2019.
Available at http://www.hermit-reasoner.com/ (accessed
Apr 2023).
[79] F. Baader, D. Calvanese, D. L. McGuinness, D. Nardi, P. F.
Patel-Schneider (Eds.), The Description Logic Handbook, 2nd
ed., Cambridge University Press, Cambridge, UK, 2007.
[80] W. W. W. C. (W3C), OWL Web Ontology Language Overview,
2012. Available at https://www.w3.org/TR/owl-featur
es/ (accessed Dec 2023).
[81] M. Hofmeister, J. Bai, G. Brownbridge, S. Mosbach, K. F. Lee,
F. Farazi, M. Hillman, M. Agarwal, S. Ganguly, J. Akroyd,
M. Kraft, Semantic Agent Framework for Automated Flood As-
sessment Using Dynamic Knowledge Graphs, 2023. Submitted
for publication. Preprint available at https://como.ceb.cam
.ac.uk/preprints/309/.
[82] ETSI, SAREF4EHAW: an extension of SAREF for eHealth
Ageing Well domain, 2020. Available at https://saref.et
si.org/saref4ehaw/v1.1.1/ (accessed Dec 2023).
[83] J. Cuenca, F. Larrinaga, E. Curry, A Unified Semantic Ontol-
ogy for Energy Management Applications, in: WSP/WOMo-
CoE@ISWC, 2017. URL: https://api.semanticschola
r.org/CorpusID:19853342.
[84] J. Cuenca, F. Larrinaga, E. Curry, DABGEO: A reusable
and usable global energy ontology for the energy do-
main, Journal of Web Semantics 61–62 (2020) 100550.
doi:10.1016/j.websem.2020.100550.
[85] C. Reinisch, M. J. Kofler, W. Kastner, ThinkHome: A smart
home as digital ecosystem, in: 4th IEEE International Confer-
ence on Digital Ecosystems and Technologies, IEEE, 2010, pp.
256–261.
[86] J. Herzen, F. Lässig, S. G. Piazzetta, T. Neuer, L. Tafti, G. Raille,
T. V. Pottelbergh, M. Pasieka, A. Skrodzki, N. Huguenin,
M. Dumonal, J. Kościsz, D. Bader, F. Gusset, M. Benheddi,
C. Williamson, M. Kosinski, M. Petrik, G. Grosch, Darts: User-
Friendly Modern Machine Learning for Time Series, Journal
of Machine Learning Research 23 (2022) 1–6. URL: http:
//jmlr.org/papers/v23/21-1177.html.
[87] S. J. Taylor, B. Letham, Forecasting at scale,
The American Statistician 72 (2018) 37–45.
doi:10.1080/00031305.2017.1380080.
[88] B. Lim, S. Ö. Arık, N. Loeff, T. Pfister, Temporal Fusion
Transformers for interpretable multi-horizon time series fore-
casting, International Journal of Forecasting 37 (2021) 1748–
1764. doi:10.1016/j.ijforecast.2021.03.012.
[89] M. Hofmeister, S. Mosbach, J. Hammacher, M. Blum,
G. Röhrig, C. Dörr, V. Flegel, A. Bhave, M. Kraft,
Resource-Optimised Generation Dispatch Strategy for
District Heating Systems Using Dynamic Hierarchi-
cal Optimisation, Applied Energy 305 (2022) 117877.
doi:10.1016/j.apenergy.2021.117877.
[90] European Environment Agency, Air quality standards, 2021.
Available at https://www.eea.europa.eu/themes/air/a
ir-quality-concentrations/air-quality-standards
(accessed Aug 2023).
[91] World Health Organization, WHO global air quality guidelines,
2021. Available at https://apps.who.int/iris/bitstr
eam/handle/10665/345329/9789240034228-eng.pdf
(accessed Aug 2023).
[92] European Environment Agency, AQ eReporting - Annual statis-
tics, 2023. Available at https://eeadmz1-cws-wp-air02.
azurewebsites.net/index.php/users-corner/stati
stical-viewer/ (accessed Aug 2023).
[93] P. D. Hamer, S.-E. Walker, G. Sousa-Santos, M. Vogt, D. Vo-
Thanh, S. Lopez-Aparicio, P. Schneider, M. O. P. Ramacher,
M. Karl, The urban dispersion model EPISODE v10.0 – Part 1:
An Eulerian and sub-grid-scale air quality model and its appli-
cation in Nordic winter conditions, Geoscientific Model Devel-
opment 13 (2020) 4323–4353. doi:10.5194/gmd-13-4323-2020.
[94] S. D. Rihm, Y. R. Tan, W. Ang, M. Hofmeister, X. Deng, M. T.
25
http://www.w3.org/DesignIssues/LinkedData.html
https://doi.org/10.4018/978-1-60960-593-3.ch008
https://blazegraph.com
https://www.w3.org/TR/sparql11-overview/
https://www.w3.org/TR/sparql11-overview/
https://doi.org/10.1016/j.future.2023.10.008
https://www.w3.org/TR/owl-time/
https://www.w3.org/TR/vocab-ssn/
https://www.w3.org/TR/vocab-ssn/
https://doi.org/10.1016/j.websem.2018.06.003
https://saref.etsi.org/core/
https://doi.org/10.1007/978-3-642-29793-9_3
https://doi.org/10.1061/(asce)cp.1943-5487.0000977
https://doi.org/10.1016/j.autcon.2011.04.014
https://doi.org/10.1016/j.autcon.2017.10.004
https://doi.org/10.1016/j.scs.2019.101857
https:doi.org/10.48550/arXiv.1201.6511
https://www.ogc.org/standards/geosparql/
https://www.ogc.org/standards/geosparql/
http://www.hermit-reasoner.com/
https://www.w3.org/TR/owl-features/
https://www.w3.org/TR/owl-features/
https://como.ceb.cam.ac.uk/preprints/309/
https://como.ceb.cam.ac.uk/preprints/309/
https://saref.etsi.org/saref4ehaw/v1.1.1/
https://saref.etsi.org/saref4ehaw/v1.1.1/
https://api.semanticscholar.org/CorpusID:19853342
https://api.semanticscholar.org/CorpusID:19853342
https://doi.org/10.1016/j.websem.2020.100550
http://jmlr.org/papers/v23/21-1177.html
http://jmlr.org/papers/v23/21-1177.html
https://doi.org/10.1080/00031305.2017.1380080
https://doi.org/10.1016/j.ijforecast.2021.03.012
https://doi.org/10.1016/j.apenergy.2021.117877
https://www.eea.europa.eu/themes/air/air-quality-concentrations/air-quality-standards
https://www.eea.europa.eu/themes/air/air-quality-concentrations/air-quality-standards
https://apps.who.int/iris/bitstream/handle/10665/345329/9789240034228-eng.pdf
https://apps.who.int/iris/bitstream/handle/10665/345329/9789240034228-eng.pdf
https://eeadmz1-cws-wp-air02.azurewebsites.net/index.php/users-corner/statistical-viewer/
https://eeadmz1-cws-wp-air02.azurewebsites.net/index.php/users-corner/statistical-viewer/
https://eeadmz1-cws-wp-air02.azurewebsites.net/index.php/users-corner/statistical-viewer/
https://doi.org/10.5194/gmd-13-4323-2020
Laksana, H. Y. Quek, J. Bai, L. Pascazio, S. C. Siong, J. Akroyd,
S. Mosbach, M. Kraft, The Digital Lab Manager: Automat-
ing Research Support, 2024. Submitted for publication. Preprint
available at https://como.ceb.cam.ac.uk/preprints/3
18/.
[95] X. Zhou, S. Zhang, M. Agarwal, J. Akroyd, S. Mosbach,
M. Kraft, Marie and BERT - A Knowledge Graph Embedding
Based Question Answering System for Chemistry, ACS Omega
8 (2023) 33039–33057. doi:10.1021/acsomega.3c05114.
[96] Ministerium für Klimaschutz, Umwelt, Energie und Mobilität,
Landesweites Solarkataster Rheinland-Pfalz, 2023. Available at
https://solarkataster.rlp.de/ (accessed Jan 2024).
[97] Hotmaps project, Hotmaps, 2020. Available at https://www.
hotmaps.eu/map (accessed Dec 2023).
[98] EEW Energy from Waste Saarbrücken GmbH, Emissionen des
Müllheizkraftwerkes Pirmasens, 2023. Available at https://
www.eew-energyfromwaste.com/fileadmin/content/
infomaterial/emissionsdaten/pirmasens/2023-03-
22_veroeffentlichung_emissionen_mhkw_pir_2022.
pdf (accessed Jan 2024).
26
https://como.ceb.cam.ac.uk/preprints/318/
https://como.ceb.cam.ac.uk/preprints/318/
https://doi.org/10.1021/acsomega.3c05114
https://solarkataster.rlp.de/
https://www.hotmaps.eu/map
https://www.hotmaps.eu/map
https://www.eew-energyfromwaste.com/fileadmin/content/infomaterial/emissionsdaten/pirmasens/2023-03-22_veroeffentlichung_emissionen_mhkw_pir_2022.pdf
https://www.eew-energyfromwaste.com/fileadmin/content/infomaterial/emissionsdaten/pirmasens/2023-03-22_veroeffentlichung_emissionen_mhkw_pir_2022.pdf
https://www.eew-energyfromwaste.com/fileadmin/content/infomaterial/emissionsdaten/pirmasens/2023-03-22_veroeffentlichung_emissionen_mhkw_pir_2022.pdf
https://www.eew-energyfromwaste.com/fileadmin/content/infomaterial/emissionsdaten/pirmasens/2023-03-22_veroeffentlichung_emissionen_mhkw_pir_2022.pdf
https://www.eew-energyfromwaste.com/fileadmin/content/infomaterial/emissionsdaten/pirmasens/2023-03-22_veroeffentlichung_emissionen_mhkw_pir_2022.pdf
Introduction
Background
Energy system modelling for smart cities
The City Energy Analyst
Interoperability gaps
Dispersion modelling
The World Avatar dynamic knowledge graph
Existing ontologies
Implementation
Ontologies
Time series ontology
District heating network ontology
Dispersion ontology
Building energy ontology
Agents
Forecasting agent
District heating optimisation agent
Emission estimation agent
Dispersion modelling agent
City Energy Analyst agent
Linked agents for cross-domain interoperability
Comprehensive energy perspective
Resource-efficient heat provision
City energy analyses and scenario planning
Impact on air quality
Conclusions
Nomenclature
Namespaces